Real-time comprehensive quality evaluation method and adaptive optical parameter adjustment system for outdoor iris recognition scene

Abstract

This invention relates to outdoor iris image processing technology, providing a real-time comprehensive quality assessment method and an adaptive optical parameter adjustment system. The assessment method includes: locating the eye boundary based on grayscale projection and local normalization; generating dual thresholds using an improved Otsu algorithm and dynamic grayscale subspace search; calculating a sharpness index based on second-order gradient entropy change; locating the pupil through a reliable edge point fitting mechanism, detecting the outer circle of the iris using an improved integral-differential operator, and evaluating segmentation accuracy using the ratio of the local standard deviation of the edge to the standard deviation of the target region; fusing gradient fitting and grayscale distribution to achieve eyelid and eyelash detection; and finally outputting a quality score based on 18 weighted indicators. The optical adjustment system senses the environment and initializes optical parameters, dynamically adjusting parameters based on weighted voting from multiple frames of images, and achieving rapid convergence with environmental consistency verification. This invention improves the speed and stability of pupil localization, sharpness assessment, eyelash detection, and illumination adjustment in low-computing-power scenarios.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of image processing technology, and particularly relates to a real-time comprehensive quality assessment method and an adaptive optical parameter adjustment system for outdoor iris recognition scenarios. Background Technology

[0002] Iris recognition, as a high-precision, non-contact biometric identification technology, has been widely used in recent years. However, its recognition performance is highly dependent on the quality of the acquired iris image. In uncontrolled environments such as outdoors, access control systems with drastic lighting changes, and mobile devices, the quality of the acquired image will be significantly reduced, mainly manifested in the following ways:

[0003] Insufficient exposure: When the user is at a distance, the infrared illumination is insufficient, resulting in reduced image contrast and difficulty in distinguishing iris texture.

[0004] Overexposure or strong ambient light interference: Strong infrared radiation in outdoor natural light can cause a sharp drop in the imaging signal-to-noise ratio, resulting in problems such as local or overall overexposure, loss of some details, and excessive pupil constriction.

[0005] Changes in user posture and distance: Users may tilt their heads, squint, blink, or move quickly, resulting in motion blur, out-of-focus blur, or occlusion of the iris area in the image.

[0006] Embedded platform computing power limitations: Mobile or embedded devices have limited computing resources and extremely high requirements for the real-time performance and efficiency of algorithms, making it difficult to run complex image quality assessment or iris preprocessing algorithms.

[0007] Existing methods for assessing iris image quality have the following limitations:

[0008] Insufficient comprehensiveness of indicators: Most methods only focus on a few basic indicators such as contrast, sharpness, focus, and usable area, lacking the evaluation of iris image-specific elements (such as pupil / iris segmentation accuracy), and cannot comprehensively and accurately determine the specific reasons for the decline in image quality.

[0009] The key challenge lies in balancing speed and accuracy: Iris quality assessment requires precise segmentation of the inner and outer boundaries of the iris, but traditional localization algorithms (such as integral-differential operators or Hough transforms) have a time complexity that is extremely high. However, its real-time performance is insufficient. Furthermore, it suffers from a high failure rate in localization on low-quality images with uneven lighting, low contrast, pupil dilation, and noise occlusion, leading to the failure of the quality assessment process. Regarding sharpness assessment, feature extraction-based methods are computationally complex and difficult to adapt to low-computing-power scenarios, such as Histogram of Oriented Gradient (HOG) features and Gabor filters. While lightweight methods based on gradient operators or frequency domain analysis are computationally efficient, their evaluation results are easily affected by differences in iris appearance. For example, a blurred image with rich texture or dense eyelashes may have a higher sharpness score than a sharp image with flat texture and sparse eyelashes, causing the system to misjudge.

[0010] Lack of effective dynamic dimming strategies: Existing iris acquisition devices mostly use static optical parameters (such as LED brightness, exposure time, and sensor gain), or have relatively simple adjustment strategies, which cannot adapt to complex and changing lighting environments. At the same time, due to the lack of a closed-loop feedback mechanism in the adjustment process, the system cannot perform fine-grained iterations based on the acquisition results, resulting in low iris image quality and affecting subsequent recognition accuracy.

[0011] Therefore, there is an urgent need to develop a computationally efficient, comprehensive iris image quality assessment algorithm that is robust to low-quality images, and to design an adaptive dynamic dimming strategy based on this algorithm. This aims to achieve adaptive iris image acquisition under complex lighting conditions on embedded platforms with limited computing power, thereby improving the stability and reliability of the iris recognition system. This invention aims to solve the aforementioned technical challenges. Summary of the Invention

[0012] To address the problems existing in current technologies, this invention provides a real-time comprehensive quality assessment method and an adaptive optics parameter adjustment system for outdoor iris recognition scenarios. This method enables rapid quality assessment of iris images in complex scenarios such as low light and strong light on embedded low-computing-power devices, and dynamically adjusts imaging parameters based on the assessment results, significantly improving the robustness and user experience of the iris recognition system.

[0013] This invention is implemented as follows: A method for real-time comprehensive quality assessment of iris images includes the following steps:

[0014] S1: Input the original iris image and reduce its size to decrease computation.

[0015] S2: Starting from the image center, search for the eye boundary in four directions (up, down, left, and right) through grayscale projection and local normalization, and calculate the position of minimum vertical integral.

[0016] S3: The Otsu binarization algorithm (OSTU) is used to dynamically binarize the eye region to generate a high threshold (to distinguish between skin and eyes). An improved Otsu binarization algorithm combined with gray sub-interval dynamic search technology is used to generate a low threshold (to distinguish between pupil and iris).

[0017] S4: Perform multi-dimensional index calculations, and statistically analyze the mean gray level, standard deviation, and medium gray level information entropy of the eye region (medium gray level is less than the high threshold of step S3 and greater than the low threshold of step S3); perform mean filtering on the eye region and execute a self-referenced sharpness evaluation algorithm based on second-order gradient entropy change.

[0018] S5: Based on the low-threshold binarization result, the pupil center is coarsely located. A search direction with the least noise is selected, and reliable pupil edge points are searched from the outside to the inside. The pupil parameters are determined by the least squares fitting method and the geometric relationship of the circle.

[0019] S6: Based on the pupil circle parameters, roughly estimate the mean of the iris region and use an improved integral-differential operator based on grayscale range to search the outer circle of the iris.

[0020] S7: Calculate the vertical gradient map of the eye region, find the reliable edge with the largest gradient in each column, and fit a quadratic function-type eyelid curve using the least squares method. Adaptively calculate the location of eyelash noise by utilizing the gray-level differences and connected component relationships between the upper and lower iris regions.

[0021] S8: Extract 14 indicators, including pupil-iris contrast, sclera-iris contrast (divided into bright and dark sides), iris texture richness, iris radius, effective iris area ratio, pupil-iris radius ratio, iris grayscale mean, eyelid opening and closing, segmentation accuracy (divided into pupil and outer iris circle), strabismus (divided into left strabismus and right strabismus), and eye pose singularity. Combine these with the four indicators from step S4: eye region grayscale mean, standard deviation, medium grayscale information entropy, and eye clarity. Normalize and weight the indicators according to preset weights, and output a comprehensive quality score.

[0022] Furthermore, the specific steps of S1 include:

[0023] Bilinear interpolation is used to scale down the original iris image, reducing the computational complexity of subsequent processing.

[0024] Furthermore, the specific steps of S2 include:

[0025] Calculate the horizontal and vertical grayscale projections of the scaled image; using the image center point as a reference, search for the local maximum values ​​of the grayscale projections of each partition along the four directions of top, bottom, left, and right; using the projection value corresponding to the center point as a reference value, and combining the local maximum values ​​in each direction, perform partitioned linear normalization on the horizontal and vertical projection curves; search for positions that satisfy the projection gradient jump condition from the center point along the four directions, determine the eye boundary position, and calculate the minimum value point of the vertical grayscale integral within the eye region as the vertical reference position of the pupil.

[0026] Furthermore, the specific steps of S3 include:

[0027] The grayscale histogram of the eye region was statistically analyzed, and a high threshold for distinguishing between skin and iris was generated using the Otsu binarization algorithm. For grayscale regions below the high threshold, a dynamic interval search strategy was adopted to divide the grayscale range into multiple sub-intervals and iterate them sequentially. An improved Otsu binarization algorithm was then used to determine the low threshold for distinguishing between pupil and iris.

[0028] Furthermore, the specific steps of S4 include:

[0029] The mean, standard deviation, and medium gray-level information entropy of the eye region are calculated based on the gray-level histogram of the eye region (medium gray level is less than the high threshold in step S3 and greater than the low threshold in step S3). For image sharpness, the medium gray-level second-order gradient distribution histograms of the original image and the image after mean filtering are calculated respectively. The information entropy of the two histograms is calculated after rounding the gradient values. The relative change in information entropy is used as a quantification of sharpness.

[0030] Furthermore, the specific steps of S5 include:

[0031] Based on the low-threshold binarized image in step S3, the minimum vertical integral value obtained in step S2 is used as the initial point. The mean value of low grayscale pixel positions is iteratively calculated in a certain neighborhood to achieve coarse localization of the pupil center. Radial search is performed in multiple directions from this point as the center, and the direction with the least noise is selected as the baseline. The pupil boundary points are detected from the outside to the inside along the baseline. The boundary points must meet three conditions at the same time: they are at the position of brightness change, the gradient of the point in the original image exceeds the predetermined threshold, and the vertical distance from the baseline increases compared with the previous boundary point. Finally, the pupil circle parameters are determined by least squares fitting and geometric constraints.

[0032] Furthermore, the specific steps of S6 include:

[0033] The average gray level of the iris is estimated by expanding the annular region based on the pupil circle parameters. The outer circle of the iris is searched within a preset radius with the pupil center as the reference. The gray level intervals of the iris and sclera are defined according to the low threshold, the average gray level of the iris, and the high threshold. The gray level integral values ​​of the candidate circles in the iris interval and the sclera interval are calculated respectively. The parameters of the outer circle of the iris are determined by maximizing the joint objective function of the two integral values.

[0034] Furthermore, the specific steps of S7 include:

[0035] For eyelid detection, the vertical gradient map of the eye region is calculated and binarized. Candidate edge points of the upper and lower eyelids are extracted, and a quadratic curve-shaped eyelid is fitted using least squares. Candidate points must meet three constraints: gray level below the high threshold in step S3, sufficient density of neighboring edge points, and smooth connectivity with adjacent points. For eyelash detection, the iris region is divided into upper and lower sub-regions. The cumulative distribution of local standard deviations at different gray levels in each sub-region is statistically analyzed. The segmentation thresholds for significant and weak eyelashes are determined based on the proportion of the cumulative distribution of the upper sub-region within the complete region. For weak eyelashes with gray levels close to the background, their topological connectivity with significant eyelashes is used for auxiliary identification.

[0036] Furthermore, the specific steps of S8 include:

[0037] Based on the aforementioned localization results, structural indicators such as pupil-iris contrast, sclera-iris contrast (divided into bright and dark sides), iris texture richness (gray-level co-occurrence matrix entropy), and effective iris area ratio were statistically analyzed. For iris texture richness, the gray-level co-occurrence matrix information entropy of the effective iris annular region was used for evaluation. For pupil-iris segmentation accuracy, the ratio of the sum of local standard deviations of the segmentation edges to the overall standard deviation of the target region was used for evaluation. The degree of strabismus was evaluated using the ratio of the distance from the pupil to the corners of the eyes. For eye pose singularity, the histogram of oriented gradients (HOG) features of the eye region were extracted, and an abnormal state such as vertical strabismus and squinting was determined by a support vector machine classifier. The classification output was then mapped to a pose abnormality probability score using a logistic regression function.

[0038] The adaptive optics parameter adjustment system based on real-time comprehensive quality assessment proposed in this invention includes the following steps:

[0039] S11: Acquire environmental images under the condition of turning off infrared LEDs, use bilinear interpolation to scale down the original face image and calculate the grayscale mean, and determine the current environment based on the grayscale mean.

[0040] S12: Set the initial values ​​of LED current, exposure time, and sensor gain based on environmental parameters.

[0041] S13: Acquire several consecutive frames of iris images and input them into the comprehensive quality assessment module. If a qualified iris image is acquired, the automatic exposure parameter adjustment process is exited and the image is output; if the image quality is unqualified, the image quality feedback adjustment stage is entered; if no iris image is detected, a full-range exposure parameter search is performed.

[0042] S14: Select indicators that are strongly correlated with brightness (mean gray value of eye area, pupil-iris contrast ratio, mean gray value of iris), and combine them with statistical models to determine the exposure direction and specific exposure parameter adjustment values.

[0043] S15: Check environmental consistency. If the environment has not changed, update the exposure parameters according to the result of S13 and jump to step S13; if the environment has changed, reinitialize the environmental parameters and jump to step S12.

[0044] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, causing the processor to perform the steps of the real-time comprehensive quality assessment method for iris images and the adaptive optics parameter adjustment system.

[0045] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the real-time comprehensive quality assessment method for iris images and the adaptive optics parameter adjustment system.

[0046] Another objective of this invention is to provide an information data processing terminal for implementing the aforementioned iris image real-time comprehensive quality assessment method and adaptive optics parameter adjustment system.

[0047] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0048] This invention proposes a fast pupil localization method based on reliable edge point fitting. Compared to classic iris segmentation methods, the proposed method achieves a dual breakthrough in computational efficiency and anti-interference capability. Through an innovative reliable edge point dynamic search and fitting mechanism, this method reduces the average time complexity of pupil localization to... The processing speed is improved by more than 5 times, and the segmentation time on low-computing-power platforms is reduced to less than 20ms. In terms of algorithm implementation, dynamic binarization and iterative grayscale compensation techniques effectively eliminate interference from light spot reflections and grayscale anomalies. Combined with a boundary point selection mechanism, only 5 to 8 effective edge points are needed to complete accurate pupil parameter fitting, which has strong adaptability to scenes with drastic pupil dilation and contraction. Test results show that in extreme scenes such as strong reflections from glasses, low illumination, and strong outdoor light, the segmentation accuracy of this algorithm remains stable at over 95%, overcoming the shortcomings of traditional gradient detection methods such as sensitivity to noise and easy failure due to insufficient contrast.

[0049] This invention proposes a self-referenced image sharpness assessment method based on medium grayscale second-order gradient entropy change. This method utilizes a customized second-order gradient operator to capture image details and uses medium grayscale constraints to focus the algorithm on the target grayscale range of the iris. Its core mechanism lies in comparing the gradient distribution information entropy changes of the original and filtered images. By employing a self-reference approach, it eliminates the interference of individual appearance differences on threshold setting, exhibiting strong robustness to motion blur and defocus blur while maintaining extremely low computational overhead. This method overcomes the problem of poor adaptability of traditional no-reference evaluation methods such as Brenner, Tenengrad, and frequency domain analysis in practical iris applications. It can provide a reliable basis for image quality pre-screening and provide technical support for the large-scale application of iris recognition in wearable devices, mobile payments, and other fields.

[0050] This invention proposes a dynamic contrast-based method for evaluating the accuracy of iris localization. This method quantifies the statistical relationship between boundary features and texture distribution by dynamically comparing the local standard deviation of the segmentation edge with the overall standard deviation of the region, thus achieving sensitive detection of segmentation anomalies. Its core logic is as follows: when segmentation is accurate, the edge gradient is large and the internal texture is uniform, resulting in a high ratio; if the radius is small or the center is offset, the edge gradient decreases, leading to a smaller ratio; if the radius is large, the overall standard deviation increases due to the inclusion of iris or scleral texture, further reducing the ratio. Testing shows that this method achieves a 97% accuracy rate in detecting segmentation anomalies without ground truth labels, effectively improving the efficiency of low-quality image interception in high-risk scenarios such as finance and security, and enhancing the reliability and security of the system.

[0051] The improved Otsu algorithm of this invention employs a dual-threshold collaborative mechanism and a dynamic search technique for grayscale sub-intervals. Test results show that this technique maintains an accuracy rate of over 99% for pupil grayscale segmentation in abnormal grayscale images such as those with insufficient infrared illumination or outdoor overexposure. It solves the failure problem of the traditional single fixed threshold method under extreme lighting conditions, which helps to expand the application of iris recognition in extreme lighting environments and enables stable deployment of iris recognition in special lighting environments such as field operations, underground operations, and disaster relief.

[0052] This invention proposes a low-contrast eyelash detection method based on multi-feature collaboration. By integrating grayscale distribution, spatial location, local contrast, and topological connectivity features, this method overcomes the problems of low detection rates or detection failures in traditional eyelash detection methods in scenarios such as excessively dark, overexposed, or absent eyelashes. It improves the detection rate of low-contrast eyelashes to 80%, significantly outperforming traditional methods based on morphological filtering. This method helps improve the integrity of biometric features in the iris region, providing accurate interference cancellation capabilities for iris recognition scenarios such as medical monitoring and border security checks, further enhancing the accuracy of iris recognition systems.

[0053] The algorithm proposed in this invention is implemented using a lightweight traditional algorithm throughout the entire process, and the processing flow is optimized. On a dual-core MIPS architecture platform (@1.2GHz), the calculation of 18 indicators takes less than 50ms and the memory usage is less than 2MB. This meets the deployment requirements of resource-constrained devices such as mobile terminals and access control systems, and helps to expand the application of iris recognition in low-computing-power scenarios, providing a safe and reliable biometric solution for the consumer electronics field.

[0054] The exposure adjustment mechanism based on multi-index voting decision proposed in this invention, combined with environmental consistency verification, enables parameter optimization to converge within 3 iterations, thereby increasing the effective image acquisition success rate of the iris recognition module to over 90% under complex lighting conditions. This helps to shorten the average recognition waiting time and improve the user experience in extreme environments. Attached Figure Description

[0055] Figure 1 This is a diagram of the real-time comprehensive quality assessment method for iris images and the adaptive optics parameter adjustment system provided in this embodiment of the invention;

[0056] Figure 2 This is a flowchart of the real-time comprehensive quality assessment method for iris images provided in this embodiment of the invention;

[0057] Figure 3 This is a flowchart of the adaptive optics parameter adjustment system based on the real-time comprehensive quality assessment method of iris images provided in this embodiment of the invention;

[0058] Figure 4 This is a flowchart of the dynamic binarization algorithm in the real-time comprehensive quality assessment method for iris images provided in this embodiment of the invention;

[0059] Figure 5 This is a flowchart of the image sharpness evaluation algorithm in the real-time comprehensive quality evaluation method for iris images provided in this embodiment of the invention;

[0060] Figure 6 This is a flowchart of the pupil localization algorithm in the real-time comprehensive quality assessment method for iris images provided in this embodiment of the invention;

[0061] Figure 7 This is a flowchart of the eyelash noise detection algorithm in the real-time comprehensive quality assessment method for iris images provided in this embodiment of the invention;

[0062] Figure 8 This is a flowchart of the pupil-iris segmentation accuracy evaluation algorithm in the real-time comprehensive quality evaluation method for iris images provided in this embodiment of the invention;

[0063] Figure 9 This is a schematic diagram of the indoor and outdoor iris acquisition effect based on static optical parameters provided in an embodiment of the present invention;

[0064] Figure 10 This is a schematic diagram illustrating the indoor iris recognition effect based on the present invention, provided in an embodiment of the present invention.

[0065] Figure 11 This is a schematic diagram illustrating the outdoor iris recognition effect based on the present invention, provided in an embodiment of the invention.

[0066] Figure 12 This is a distribution map of the clarity index of self-collected data provided in an embodiment of the present invention;

[0067] Figure 13 This is a schematic diagram of the pupil positioning effect provided in an embodiment of the present invention;

[0068] Figure 14 This is a distribution chart of pupil segmentation accuracy indicators provided in an embodiment of the present invention;

[0069] Figure 15 This is a schematic diagram of the eyelash detection effect provided in an embodiment of the present invention;

[0070] Figure 16 This is an iris comprehensive quality score distribution map provided in an embodiment of the present invention;

[0071] Figure 17 This is a schematic diagram of the comprehensive iris quality score provided in an embodiment of the present invention. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0073] like Figure 1 As shown in the figure, the real-time comprehensive quality assessment method and adaptive optical parameter adjustment system for outdoor iris recognition scenarios provided by the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0074] The following is a specific example demonstrating how to apply a real-time comprehensive quality assessment method and an adaptive optics parameter adjustment system for outdoor iris recognition scenarios.

[0075] Example: Tablet PC Iris Recognition Module

[0076] In tablet-based iris recognition scenarios, iris recognition systems face challenges such as device shake, insufficient exposure, complex ambient lighting, and low user cooperation.

[0077] The real-time comprehensive quality assessment method for iris images provided in this embodiment of the invention is implemented as follows:

[0078] 1. Input Image: The original iris image is scaled to reduce computational load.

[0079] 2. Eye Region Range Search: Execute the eye region search algorithm to determine the upper, lower, left, and right boundaries of the eye region, and at the same time coarsely locate the vertical coordinates of the pupil region.

[0080] 3. Dynamic Dual Threshold Binarization: Execute the dynamic dual threshold binarization algorithm for the iris image to determine the high threshold for distinguishing between the eyes and skin, and the low threshold for distinguishing between the pupil and the iris.

[0081] 4. Multi-dimensional index calculation: Calculate the mean gray value, standard deviation, and medium gray value information entropy of the eye region (medium gray value is less than the high threshold in step 3 and greater than the low threshold in step 3), and execute the self-reference sharpness algorithm based on the second-order gradient entropy change of medium gray value to perform coarse screening of iris image quality.

[0082] 5. Rapid pupil localization: Execute a pupil localization algorithm based on reliable edge point search and fitting to determine the pupil center coordinates and radius.

[0083] 6. Iris outer edge localization: Execute an improved grayscale range-based integral-differential operator to determine the center coordinates and radius of the outer circle of the iris.

[0084] 7. Eyelid and eyelash detection: Execute the eyelid localization algorithm and the multi-feature eyelash detection algorithm to determine the eyelid curve parameters and the eyelash noise mask.

[0085] 8. Comprehensive index extraction and evaluation: The image comprehensive quality evaluation algorithm is executed to extract 18 indicators, including pupil-iris contrast, sclera-iris contrast (divided into bright and dark sides), iris texture richness, pose singularity, etc. The indicators are normalized and weighted according to preset weights to output a comprehensive quality score.

[0086] The adaptive optics parameter adjustment system based on real-time comprehensive quality assessment provided in this embodiment of the invention is implemented as follows:

[0087] 1. Initial Environment Detection: Execute the initial environment detection method, acquire environmental images with infrared LEDs turned off, and determine the environment based on the average value of the images.

[0088] 2. Optical parameter settings: Set the LED current, exposure time, and initial sensor gain value according to environmental parameters.

[0089] 3. Image Quality Assessment: Acquire several consecutive frames of iris images and input them into the comprehensive quality assessment module. If a qualified iris image is acquired, the automatic exposure parameter adjustment process ends and the image is output; if the image quality is unqualified, the image quality feedback adjustment stage begins; if no iris image is detected, a full-range exposure parameter search is performed.

[0090] 4. Exposure decision: Select indicators that are strongly correlated with brightness, and combine them with statistical models to determine the exposure direction and specific exposure parameter adjustment values.

[0091] 5. Environmental consistency check: If the environment has not changed, update the exposure parameters according to the result of step 4; otherwise, reinitialize the environmental parameters.

[0092] This implementation example demonstrates the applicability of the real-time comprehensive quality assessment method for iris images and the adaptive optics parameter adjustment system of the present invention to low-quality iris image acquisition scenarios. It can effectively identify image quality problems caused by equipment jitter, insufficient exposure, complex ambient lighting, and low user cooperation during iris acquisition, which helps to improve the reliability of the iris recognition system and expand its application scenarios.

[0093] like Figure 1 As shown, the image processing procedure of the real-time comprehensive quality assessment method for outdoor iris images involves multiple steps. The aim is to improve the efficiency and reliability of the iris recognition system by optimizing the algorithm flow and performing comprehensive quality assessment of iris images from multiple dimensions. The following is a detailed description of each step in the real-time comprehensive quality assessment method for iris images:

[0094] Step S1: Image scaling

[0095] Bilinear interpolation is used to reduce the size of the original iris image, thereby reducing the computational complexity of subsequent processing.

[0096] Step S2: Eye area search

[0097] Calculate grayscale projection: Calculate the horizontal and vertical grayscale projections of the scaled image.

[0098] Finding local maxima: Using the image center point as a reference, search for the local maxima of grayscale projection in each partition along the four directions of top, bottom, left, and right.

[0099] Partitioned linear normalization: Using the projection value corresponding to the center point as a reference value, and combining the local maximum values ​​in each direction, partitioned linear normalization is performed on the horizontal and vertical projection curves.

[0100] Eye boundary search: Search along the four directions from the center point for positions that satisfy the projection gradient jump condition, and determine them as eye boundary positions.

[0101] Pupil Vertical Reference Position Search: Based on the eye region boundary, search for the point with the minimum vertical grayscale integral as the pupil vertical reference position.

[0102] Step S3: Dynamic dual-threshold binarization

[0103] Gray-level histogram statistics: Statistical analysis of the gray-level histogram of the eye region.

[0104] High threshold search: On the complete eye region, a high threshold is generated to distinguish between the skin and the iris using the Otsu binarization algorithm.

[0105] Dynamic search of grayscale sub-intervals: For low grayscale regions below the high threshold, a dynamic interval search strategy is adopted to divide the grayscale range into multiple sub-intervals and iterate them sequentially. The low threshold for distinguishing the pupil and iris is determined by the improved Otsu binarization algorithm.

[0106] Step S4: Calculation of multi-dimensional indicators

[0107] Multi-dimensional grayscale index calculation: Calculate the mean grayscale value, standard deviation, and medium grayscale information entropy of the eye area.

[0108] Sharpness index calculation: Calculate the second-order gradient value distribution histograms of the medium gray level (less than the high threshold in step S2 and greater than the low threshold in step S3) of the original image and the mean-filtered image respectively. After rounding the gradient values, calculate the information entropy of the two histograms. Use the relative change in information entropy as the sharpness metric.

[0109] Step S5: Rapid Pupil Localization

[0110] Coarse localization of the pupil center: Based on the low-threshold binarization result, the minimum vertical integral obtained by S2 is used as the initial vertical coordinate, and the image center is used as the initial horizontal coordinate. The mean position of low grayscale pixels in its neighborhood is calculated iteratively to achieve coarse localization of the pupil center.

[0111] Determine the search baseline: Using the coarse pupil localization result as the center, conduct a multi-directional radial search in the low-threshold binarized image, and select the direction with the least noise as the baseline.

[0112] Pupil boundary search: Detect pupil boundary points from the outside to the inside along the baseline. The boundary point must meet three conditions at the same time: it is located at a position of brightness change, the gradient of the point in the original image exceeds a predetermined threshold, and the vertical distance from the baseline increases compared to the previous boundary point.

[0113] Pupil parameter fitting: Pupil circle parameters are determined based on least squares fitting and geometric constraints of the circle.

[0114] Step S6: Iris outer edge positioning

[0115] Iris gray mean estimation: The average gray value of the iris is estimated by expanding the annular region based on the pupil circle parameter.

[0116] Iris boundary search: Define the gray range of the iris and sclera based on the low threshold, the mean gray value of the iris, and the high threshold. Calculate the gray integral value of the candidate circle in the iris range and the sclera range respectively. Determine the parameters of the outer circle of the iris by maximizing the joint objective function of the two integral values.

[0117] Step S7: Eyelid and eyelash inspection

[0118] Eyelid noise detection: Calculate and binarize the vertical gradient map of the eye region, extract candidate edge points for the upper and lower eyelids, and fit a quadratic curve shape of the eyelid using least squares. Candidate points must meet three constraints: gray level below a high threshold, sufficient density of neighboring edge points, and smooth connectivity with adjacent points.

[0119] Eyelash noise detection: The iris region is divided into upper and lower sub-regions. The cumulative distribution of local standard deviations of different gray levels in each sub-region is statistically analyzed. The significant eyelash segmentation threshold and weak eyelash segmentation threshold are determined based on the proportion of the cumulative distribution of the upper sub-region in the complete region. For weak eyelashes with gray levels close to the background, their topological connectivity with significant eyelashes is used for auxiliary identification.

[0120] Step S8: Comprehensive Indicator Extraction and Evaluation

[0121] Extraction metrics: 18 metrics were extracted, including pupil-iris contrast, sclera-iris contrast (separating the bright and dark sides), iris texture richness, effective iris area ratio, eyelid opening and closing, segmentation accuracy (separating the pupil and iris), and pose strangeness.

[0122] Indicator normalization and fusion: The indicators are normalized and weighted according to preset weights to output a comprehensive quality score.

[0123] Through the detailed image processing described above, this method can evaluate iris image quality in real time, reduce the risk of low-quality images entering the recognition process, and improve the reliability and applicability of the iris recognition system.

[0124] See Figure 1 The adaptive optics parameter adjustment system provided in this invention uses a closed-loop feedback control mechanism based on a real-time comprehensive quality assessment method for iris images. The system structure is designed as follows: an environment detection module initializes environmental parameters, sets initial optical parameter values ​​based on these parameters, acquires several consecutive frames of iris images, inputs them to the comprehensive quality assessment module, an exposure decision module determines the exposure direction and specific exposure parameter adjustment values, and an environment consistency detection module determines whether to perform feedback adjustment. The final output of the system is a qualified iris image. The following is a detailed description of each step in the adaptive optics parameter adjustment system based on the real-time comprehensive quality assessment method for iris images:

[0125] S1: Environment Initialization

[0126] Environmental images are acquired with infrared LEDs turned off. The original facial images are scaled down using bilinear interpolation and the mean grayscale value is calculated. The current environment is determined based on the mean grayscale value.

[0127] S2: Optical parameter initialization

[0128] Set the LED current, exposure time, and initial sensor gain based on environmental parameters.

[0129] S3: Image Acquisition and Overall Quality Assessment

[0130] A series of consecutive iris images are acquired and input into the comprehensive quality assessment module for evaluation. If a qualified iris image is acquired, the automatic exposure parameter adjustment process is exited and the image is output; if the image quality is unqualified, the image quality feedback adjustment stage is entered; if no iris image is detected, a full-range exposure parameter search is performed.

[0131] S4: Exposure Decision

[0132] Select indicators that are strongly correlated with brightness (mean gray value of eye area, pupil-iris contrast ratio, mean gray value of iris), and combine them with statistical models to determine the exposure direction and specific exposure parameter adjustment values.

[0133] S5: Environmental Consistency Testing

[0134] To verify environmental consistency, if the environment has not changed, update the exposure parameters based on the exposure decision results and re-execute image acquisition and comprehensive quality assessment; if the environment has changed, reinitialize the environmental parameters.

[0135] like Figure 2 As shown, the real-time comprehensive quality assessment method for outdoor iris recognition provided in this embodiment of the invention specifically includes the following steps:

[0136] S1. To improve the processing speed of the real-time comprehensive quality assessment method for iris images, the input original iris image needs to be scaled. Reducing the image size can effectively reduce the computational load of subsequent processes and ensure the real-time performance of the comprehensive iris quality assessment;

[0137] Step S1 specifically involves using bilinear interpolation to reduce the height and width of the original iris image to 0.4 times its original size. The reduced image is denoted by I, with a height of H and a width of W.

[0138] S2. Iris images may contain a large number of low grayscale noise areas, such as eyebrows, hair, and black eyeglass frames, which can interfere with subsequent processing. Therefore, it is necessary to determine the boundary of the eye region in the image to narrow the search range of subsequent algorithms.

[0139] Step S2 specifically involves first calculating the grayscale projection HS in the horizontal direction and the projection VS in the vertical direction of image I, where i represents the row of a pixel and j represents the column of a pixel. The projection calculation formula is as follows:

[0140]

[0141] Using the image center point as a reference, the local maximum value of the grayscale projection of each partition is searched along the four directions of top, bottom, left, and right. The maximum value of the horizontal projection of the upper half of the image is used as... This indicates that the maximum horizontal projection value of the lower half is represented by... This indicates that the maximum vertical projection value of the left half is represented by... This indicates that the maximum vertical projection value of the right half is represented by... This indicates that the horizontal projection value corresponding to the center point is... The vertical projection value corresponding to the center point is Using the projection value corresponding to the center point as a reference, and combining it with the local maximum values ​​in each direction, a partitioned linear normalization is performed on the horizontal and vertical projection curves to obtain the normalized horizontal projection HSN and vertical projection VSN. The normalization process can be expressed by the following formula:

[0142]

[0143] For the normalized grayscale projection, search along the four directions from the center point for positions that satisfy the projection gradient jump condition, and determine these as the eye boundary positions. The upper boundary ordinate is used as... The lower boundary ordinate is indicated by... The left boundary x-coordinate is used The right boundary x-coordinate is represented by... The minimum vertical grayscale integral point within the boundary region is calculated as the vertical reference position of the pupil. The gradient calculation is defined using the central difference method as follows:

[0144]

[0145] from Begin a decreasing search until the first condition is met:

[0146]

[0147] Location ,in This is the threshold for gradient jumps during upward search;

[0148] from Begin an incremental search to find the first satisfied condition:

[0149]

[0150] Location ,in This is the threshold for gradient jumps during downward search;

[0151] from Begin a decreasing search until the first condition is met:

[0152]

[0153] Location ,in This is the threshold for gradient jump when searching to the left;

[0154] from Begin an incremental search to find the first satisfied condition:

[0155]

[0156] Location ,in This is the gradient jump threshold for searching to the right.

[0157] S3, see also Figure 4 Dynamic dual-threshold binarization is performed on the eye region. The high threshold is used to distinguish between the skin region and the eye region, where the gray-level difference is significant, and the ordinary Otsu thresholding segmentation method is suitable. The low threshold is used to distinguish between the pupil region and the iris region, where the gray-level difference between the two is small and there is a significant difference in gray-level distribution between indoor and outdoor images. Therefore, the gray-level range needs to be divided into multiple sub-intervals and a dynamic search strategy needs to be implemented.

[0158] Step S3 specifically involves calculating the grayscale distribution histogram of the eye region. First, the high threshold is determined using the standard Otsu thresholding method. All grayscale values ​​less than Pixels are marked as low grayscale regions. For low grayscale regions, an improved Otsu thresholding method is used to search for low threshold values. Set the initial search interval as The improved Otsu thresholding method searches for a threshold that maximizes the inter-class variance within an interval. The optimization objective is shown in the following equation:

[0159]

[0160] in:

[0161]

[0162]

[0163]

[0164] The improved optimization objective includes the proportion of the pupil area in the entire low-grayscale region. For search results, if among them... Less than the threshold Then Increasing Then search again; this usually corresponds to outdoor overexposure. Greater than the threshold Then Decreasing Then perform a re-search, which typically corresponds to insufficient indoor infrared lighting. If the conditions are met... In If the search is successful, then the search was successful.

[0165] Incorporating the pupil area ratio into the optimization objective of the Otsu threshold segmentation method effectively addresses the segmentation failure issue caused by pupil constriction and increased grayscale in outdoor scenes. Since the grayscale of the pupil is significantly higher outdoors than indoors, using the same low threshold search range as indoors can easily lead to the entire outdoor image being misclassified as a high-grayscale region. The value is 0. This is introduced into the optimization objective. This system can detect such anomalies and gradually expand the search range to the outdoor grayscale range by using a search strategy that increases from low grayscale to high grayscale. This adaptive search mechanism can not only match different scenes, but also generate a more conservative pupil segmentation threshold, thereby effectively eliminating interference from eyelashes and iris textures in the subsequent localization stage.

[0166] For threshold and The determination of the threshold range is based on the fact that the invalid segmentation area caused by outdoor overexposure is much smaller than the normal pupil area, while the invalid segmentation area caused by indoor over-darkness is significantly larger than the normal range. Therefore, this algorithm determines the threshold range... The settings have high fault tolerance, which can effectively overcome the impact of extreme environments such as overexposure and underexposure on pupil grayscale segmentation, and provide stable prior information on pupil grayscale for subsequent positioning.

[0167] S4. Before proceeding to subsequent algorithm steps, a preliminary evaluation of the iris image is required to assess its contrast and sharpness, preventing images with severely insufficient contrast and sharpness from entering the later stages. Evaluation metrics include the mean grayscale value, standard deviation, and medium grayscale entropy (medium grayscale is defined as less than the high threshold) of the eye region. And greater than the low threshold (and clarity.)

[0168] Step S4 specifically involves first calculating the mean grayscale value, standard deviation, and medium grayscale entropy of the eye region. The mean grayscale value follows a normal distribution on the dataset and is a two-sided constrained indicator; excessively dark images will have a lower mean grayscale value, while overexposed images will have a higher mean grayscale value. These need to be transformed into positive indicators through normal distribution parameter estimation. The standard deviation and medium grayscale entropy are positive indicators; both excessively dark and overexposed images will result in lower values. The calculation process for medium grayscale entropy is as follows: first, calculate the grayscale value of the eye region within the medium grayscale range (below the high threshold). And greater than the low threshold histogram of ) Then, normalization is performed, and the information entropy of the normalized histogram is calculated. The specific calculation formula is as follows:

[0169]

[0170] See Figure 5 For eye clarity, the medium gray level (less than the high threshold) in the eye region is calculated. And greater than the low threshold The second gradient of a pixel The calculation formula is shown below:

[0171]

[0172]

[0173]

[0174] in, For fixed weights, , , It is a constant. The direction of the operator.

[0175] For second-order gradients Quantization and rounding are performed, and the second-order gradient distribution histogram of all medium-grayscale pixels is plotted. The information entropy is then calculated using the histogram. Let the information entropy of the medium-grayscale second-order gradient distribution in the original image be... The entropy of the second-order gradient distribution of medium gray levels in the image after mean filtering is: The relative change in information entropy before and after mean filtering is used as a quantification of clarity, and the specific formula is as follows:

[0176]

[0177]

[0178] Among them, the clarity evaluation index The normal distribution exhibits normal distribution characteristics on the dataset and is a two-sided constrained index. Its characteristic is that image sharpness is higher when the value is closer to the center and lower when the value is closer to the sides. The normal distribution parameters can be obtained through normal distribution parameter estimation. and ,Will Convert to a positive clarity index .

[0179] Since iris texture is essentially a high-frequency detail in an image and accounts for a small proportion of the overall image energy, sharpness detection algorithms must possess extremely high sensitivity and target specificity. The method described above focuses on the effective iris region by limiting grayscale, captures subtle structural changes using second-order gradients, and decouples image content from sharpness features through a self-reference mechanism. This enables it to effectively distinguish between motion-blurred and defocused images when no sharp image is available as a reference, while also addressing the limited applicability of traditional sharpness evaluation algorithms in iris recognition applications.

[0180] Furthermore, motion blur and defocus blur can be equivalent to low-pass filtering, which can lead to damage to the iris structure and weakening of edges. Since subsequent steps such as iris localization, eyelid detection, and eyelash segmentation are highly dependent on the gradient information of the image, pre-eliminating severely blurred images through a sharpness evaluation mechanism can effectively avoid the negative impact of low-quality input on subsequent algorithms, thereby improving the overall stability of the system.

[0181] S5, see also Figure 6 Based on the aforementioned binarization results, a reliable edge point search and fitting strategy is employed to locate the pupil. The pupil position serves as the benchmark for locating the outer circle of the iris, sclera, eyelids, and eyelashes; accurate pupil localization is a crucial task in the iris preprocessing workflow. Iris acquisition systems typically use direct infrared light to illuminate the eye, which creates a bright reflective area on the pupil, compromising its integrity. In open acquisition scenarios, different environments can negatively impact pupil localization: insufficient infrared illumination reduces the contrast between the pupil and iris; when the user wears glasses, large light spots can easily obscure the pupil and iris areas; under strong outdoor light, the pupil constricts significantly and exhibits grayscale anomalies; in dark environments, the pupil dilates rapidly, resulting in abnormal shapes. These factors affect pupil detection performance in complex environments, representing a key performance bottleneck in iris recognition systems.

[0182] Step S5 specifically involves, based on a low threshold Binarize the glasses area by setting low grayscale values ​​to 0 and high grayscale values ​​to 255. The binarized image is denoted as [image name missing]. The minimum value of the vertical integral obtained in step S2. Initialize the x-coordinate of the pupil center Centered on the eye area Initialize the y-coordinate of the pupil center In its certain neighborhood The algorithm calculates the mean position of low-grayscale pixels and iterates multiple times using this mean position as the new starting point until the position changes converge. The final position is recorded as point A. In the horizontal axis direction, since the noise distribution generated by eyelashes is uniform, averaging the positions will not cause a deviation in the horizontal coordinate; therefore, its neighborhood point set... This can be expressed by the formula:

[0183]

[0184] in This is a constant. On the vertical axis, using positional averaging causes eyelash noise to skew the vertical axis coordinate upwards. The method to mask eyelash noise is to search for three transition points from bottom to top: the first transition point from dark to light, the second transition point from dark to light, and the third transition point from light to dark. The vertical coordinates of these three transition points are averaged, and low-grayscale points above the average position are removed, resulting in the neighborhood point set along the vertical axis. This can be expressed by the formula:

[0185]

[0186] in, It is the average position of the three transition points in the j-th column, and c is a constant.

[0187] Finally, the neighborhood used for position averaging It can be represented as:

[0188]

[0189] Due to factors such as reflective areas, the pupil area formed after binarization is incomplete, and the pupil center cannot be accurately located using the position mean.

[0190] With point The search radiates outwards from the center in multiple directions, selecting the direction with the least noise. As a baseline In binarized images, the area above the pupil is often accompanied by noise from eyelashes, and the upper half of the pupil may be obscured by the eyelid. Therefore, the search direction is usually directly below the pupil, directly below by a 45° clockwise rotation, and directly below by a 45° counterclockwise rotation. After determining the search baseline, pupil boundary points are detected sequentially from the outside to the inside along the baseline. A boundary point must simultaneously meet three conditions: it must be located at a brightness transition point; its gradient in the original image must exceed a predetermined threshold; and its vertical distance from the baseline must increase compared to the previous boundary point on the same side. The specific formula for selecting boundary points is as follows:

[0191]

[0192]

[0193]

[0194] where 、 、 are constants, represents the gradient magnitude of point in the original image. The gradient calculation can adopt classic image gradient calculation methods such as Sobel and Laplace operators. represents the perpendicular distance of point from the reference line Point represents the previous reliable pupil edge point on the same side of the reference line as point

[0195] After all reliable pupil edge points are selected, when there are sufficient candidate points (N > T, where N is the number of candidate points and T is a fixed threshold), the least squares method is used to fit the pupil parameters; when there are insufficient candidate points (2 < N ≤ T), the pupil center and radius are determined based on the geometric properties of the circle; if the number of candidate points is too small (N ≤ 2) or the positioning accuracy index does not meet the standard, it is determined as a failure, and the adaptive threshold iteration mechanism is triggered to correct the gray anomaly by increasing the binarization threshold, and step S5 is executed again until the positioning is successful or the iteration upper limit is reached.

[0196] The test results show that the above pupil method shows extremely high positioning efficiency and robustness for outdoor low-quality images. In terms of computational efficiency, by converting the high-dimensional search into a low-dimensional fitting, the time complexity of the core steps is reduced to O(1), and the computational time consumption is extremely low and constant. In terms of robustness, the algorithm integrates multiple anti-interference mechanisms: adaptive binarization and threshold iteration are used to eliminate specular reflection and extreme illumination effects, sparse fitting is used to adapt to drastic pupil scaling, and pure edges are extracted by searching the reference line to complete accurate pupil fitting in complex environments.

[0197] S6. According to the pupil circle parameters, roughly estimate the mean value of the iris region, and use an improved integral and differential operator based on the gray range to search for the outer circle of the iris. <00005�0>

[0198] Specifically, step S6 is to expand the annular region according to the pupil center coordinates and radius, and sample on the annular region to estimate the average gray value of the iris <00,00542,. Define the iris gray interval with the low threshold , the high threshold and the average gray value of the iris and the gray interval of the sclera​​ ,in and The definition is as follows:

[0199]

[0200] To improve processing speed, the outer circle of the iris and the pupil circle are considered as concentric circles, with the center of the pupil circle as the reference point. Using this as a baseline, the outer circle of the iris is searched within a preset radius. The grayscale integral values ​​of the candidate circles in the iris and sclera regions are calculated separately. The parameters of the outer circle of the iris are determined by maximizing the joint objective function of the two integral values. This process can be expressed as:

[0201]

[0202] in Let be the radius of the candidate circle. It is a Gaussian filter function. It is by The differential length element of the determined circle. At the outer boundary of the iris, the integrals within the iris grayscale and scleral grayscale ranges will simultaneously exhibit significant jumps, maximizing the joint objective function. This improvement enhances the robustness of the outer iris circle search process to noisy images, particularly improving the segmentation success rate for images with small eye areas and low contrast.

[0203] S7. Calculate the vertical gradient map of the eye region, find the reliable edge with the largest gradient in each column, fit a quadratic function-type eyelid curve using the least squares method, and adaptively calculate the location of eyelash noise by utilizing the gray-level difference and connected component relationship between the upper and lower regions of the iris. Eyelids and eyelashes are the two most significant types of noise in iris images, and accurately removing eyelids and eyelashes from iris images is crucial for improving iris recognition accuracy.

[0204] Step S7 specifically involves, for eyelid detection, first calculating the vertical gradient map of the eye region. And perform binarization Candidate edge points for the upper and lower eyelids are extracted. Candidate points for the upper eyelid are searched from top to bottom, and candidate points for the lower eyelid are searched from bottom to top. Candidate points must meet three constraints: grayscale below a high threshold, sufficient density of edge points within their neighborhood, and smooth connectivity with adjacent edge points. A neighborhood is defined as the set of all points directly or indirectly connected to a given point through a four-adjacency relationship. It can be expressed as follows:

[0205]

[0206] in The effective area of ​​the image. These three constraints are constants. They can be expressed by the following formula:

[0207]

[0208]

[0209]

[0210] in This represents the number of points in the set. To and The y-coordinate of the adjacent previous candidate point, and The constant is used. After obtaining all eyelid candidate points, a quadratic curve shape of the eyelid is fitted using least squares.

[0211] See Figure 7 For eyelash detection, the iris region is divided into upper and lower sub-regions. The cumulative distribution of local standard deviations for different gray levels in each sub-region is statistically analyzed. The eyelash segmentation threshold is determined based on the proportion of the cumulative distribution of the upper sub-region within the complete region. Regarding window size... Local standard deviation and local mean The calculation formula is as follows:

[0212]

[0213]

[0214] area various gray levels The formula for calculating the cumulative local standard deviation is as follows:

[0215]

[0216] Cumulative distribution function of local standard deviations for different gray levels:

[0217]

[0218] The cumulative distribution function for the complete iris region Cumulative distribution function of the upper subregion of the iris Its significant characteristic is that the two almost completely overlap in the low grayscale area, while as the grayscale gradually increases, and Gradually separated. This is because low-grayscale eyelashes mainly appear in the upper half of the iris region, while other grayscale pixels are relatively evenly distributed throughout the entire iris region. Divide by The proportion of the cumulative value of the upper sub-region to the cumulative value of the complete iris region was obtained. :

[0219]

[0220] A notable characteristic of the curve is that it rises rapidly to its maximum value in the low grayscale region, then begins to decline, and eventually stabilizes between 0.5 and 0.6. The gray level corresponding to the maximum value is used as the significant eyelash segmentation threshold, and then the search continues towards higher thresholds to find... For transition points below a certain threshold, the corresponding gray level is used as the weak eyelash segmentation threshold. For weak eyelash candidate points, topological connectivity with salient eyelashes is used for auxiliary identification. Only weak eyelash candidate points that satisfy an eight-neighbor relationship with salient eyelashes can be finally confirmed as eyelashes. This adjacency relationship is represented as follows:

[0221]

[0222] in For the point set formed by strong eyelashes, For the point set formed by weak eyelashes, This refers to the grayscale area of ​​the iris. It is a positive integer. This is the symbol for the eight-neighbor domain.

[0223] S8. Extract 14 indicators, including pupil-iris contrast, sclera-iris contrast (divided into bright and dark sides), iris texture richness, iris radius, effective iris area ratio, pupil-iris radius ratio, iris grayscale mean, eyelid opening and closing, segmentation accuracy (pupil and iris), strabismus (divided into left and right strabismus), and eye pose singularity. Combine these with the four indicators from step S4: eye region grayscale mean, standard deviation, medium grayscale information entropy, and eye clarity. Normalize and weight the indicators according to preset weights, and output a comprehensive quality score.

[0224] Step S8 specifically involves first extracting iris structural indicators based on the aforementioned iris image information, including pupil-iris contrast, sclera-iris contrast, iris radius, effective iris area ratio, pupil-iris radius ratio, iris grayscale mean, eyelid opening and closing degree, and eye strabismus. Among them, pupil-iris contrast is calculated by subtracting the mean pupil area from the mean iris area; sclera-iris contrast is calculated by subtracting the mean iris area from the mean sclera area. Since the side of the sclera closest to the bridge of the nose is usually in the area covered by the cross-coverage of the left and right infrared supplementary lights, the image brightness is higher, and the grayscale of the sclera is prone to showing the uneven grayscale of the two sides of the iris. Therefore, bright-side sclera-iris contrast and dark-side sclera-iris contrast are distinguished; effective iris area ratio refers to the proportion of the effective iris area after removing noise to the area of ​​the complete iris annular region; pupil-iris radius ratio refers to the ratio of the difference between the iris radius and the pupil radius to the iris radius, used to evaluate pupil dilation and contraction; eyelid opening and closing refers to the ratio of the average distance between the upper and lower eyelids to the iris diameter; for strabismus, the ratio of the distance from the pupil area to the two corners of the eyes is used for evaluation, and the position of the corners of the eyes is determined by eyelid candidate points.

[0225] Iris texture richness is a key biometric indicator, and its extraction requires quantifying the complexity of grayscale changes in local regions. In this embodiment of the invention, iris texture richness is evaluated by calculating the grayscale co-occurrence matrix of the effective annular region of the iris and deriving its information entropy value. Higher information entropy indicates richer and more diverse texture patterns, while lower information entropy indicates flat and sparse textures.

[0226] See Figure 8 For the accuracy of pupil and iris segmentation, the ratio of the sum of the local standard deviations of the segmentation edges to the overall standard deviation of the target region is used for evaluation. Let the window size be... The formula for calculating the local standard deviation is shown in step S7. The sum of the local standard deviations of the segmentation edges can be expressed by the following formula:

[0227]

[0228] in The iris area after removing noisy areas such as eyelids and eyelashes. Parameters of the outer circle of the pupil or iris. Overall standard deviation of the target region. The calculation formula is as follows:

[0229]

[0230]

[0231] in, The total area of ​​the target region, and the final segmentation accuracy index. It can be expressed as follows:

[0232]

[0233] This method achieves sensitive detection of segmentation anomalies by accurately quantifying the statistical distribution relationship between iris boundary features and regional texture. Taking the outer edge of the iris as an example, when the iris boundary is accurately segmented, the numerator (local standard deviation at the edge) increases significantly due to the abrupt change in boundary gradient, while the denominator (overall standard deviation of the region) remains at a low level due to the uniformity of iris texture, resulting in a high dynamic ratio. When the segmentation radius is too small or the center is offset, the local standard deviation at the boundary decreases while the overall standard deviation of the region remains basically unchanged, leading to a decrease in the ratio. When the segmentation radius is too large or the center is offset, the local standard deviation at the boundary decreases while the overall standard deviation of the region increases due to the inclusion of more scleral texture, further reducing the ratio.

[0234] To address eye pose singularities, the Histogram of Oriented Gradients (HOG) of the eye region is extracted. A support vector machine classifier is then used to determine whether the eye exhibits abnormal states such as strabismus or squinting (binary classification: normal or abnormal). Finally, a logistic regression function is used to map the classification output to a pose abnormality probability score. This process can be represented by the following formula:

[0235]

[0236]

[0237]

[0238] in Indicates the HOG feature dimension. This represents the weight vector of the support vector machine classifier. For bias terms, and These are the calibration parameters for the Sigmoid function.

[0239] After extracting all evaluation indicators, the indicators need to be normalized according to the preset weights to obtain individual scores. All indicators can be divided into two types:

[0240] The first category is positive indicators. The larger the indicator value, the higher the image quality. The corresponding indicators are: sclera-iris contrast, iris texture richness, iris radius, effective iris area ratio, eyelid opening and closing, segmentation accuracy, eye pose singularity, eye region standard deviation, and medium grayscale information entropy.

[0241] The second type is two-sided constraint indicators. The closer the value of these indicators is to the statistical mean of the dataset, the higher the image quality. The farther away the value is from the mean, the lower the image quality. Such indicators include: pupil-iris contrast, pupil-iris radius ratio, iris gray mean, strabismus, eye region gray mean, and eye sharpness.

[0242] Since the two-sided constraint indicators exhibit a normal distribution on the dataset, they need to be converted into positive indicators through normal distribution parameter estimation. The conversion process is shown in step S4 for the clarity indicator. After converting all indicators into positive indicators, a linear mapping is used to map them to... On the quality score. For each indicator, the linear mapping needs to determine a minimum reference value for that indicator. The corresponding minimum reference score Determine the maximum reference value for an indicator. The corresponding minimum reference score Then use the following formula for mapping:

[0243]

[0244] Finally, a weighted average of all individual scores is taken to output the overall quality score. This process is shown in the following formula:

[0245]

[0246] like Figure 3 As shown, the adaptive optics parameter adjustment system provided in this embodiment of the invention specifically includes the following steps:

[0247] S1. Determine the environment based on the infrared intensity in the environment.

[0248] Step S1 specifically involves acquiring an environmental image under the conditions of turning off the infrared LEDs and setting the sensor exposure time and gain to fixed values. The original facial image is then scaled down using bilinear interpolation, and the grayscale mean is calculated. The current environment is determined based on the grayscale mean. In this embodiment, the environment is divided into five categories:

[0249] (1) Controllable artificial light source-dominated scenario: The typical characteristics of this scenario are that the ambient illuminance is less than or equal to 500 lux (the unit of illuminance, symbol lux), the light source is mainly artificial lighting, there is no direct natural light, the light distribution is uniform and the dynamic change rate is low. This scenario will be referred to as in the following text. ;

[0250] (2) Ceiling-Shielded Ambient Light Mixed Scene: This scene is an indoor-outdoor transition scene. Its typical characteristics are that there is a physical shielding structure (ceiling or canopy) at the top of the environment, and the side opening leads to diffused natural light. The ambient illuminance is 500-10000 lux, presenting a non-uniform light distribution. This scene will be referred to as in the following text. ;

[0251] (3) Localized occlusion-type strong ambient light scattering scene: The typical characteristic of this scene is that under ambient light greater than 10,000 lux, localized shadows are caused by buildings or vegetation, etc. This scene will be referred to as in the following text. ;

[0252] (4) Low to medium intensity direct sunlight scenario: This scenario is characterized by oblique sunlight generated when the sun's altitude angle is less than 45°, with an ambient illuminance of 10,000-50,000 lux. Unidirectional high light reflection may occur in the iris area. This scenario will be referred to as [missing information] in the following text. ;

[0253] (5) High-intensity vertical solar radiation scene: The typical characteristics of this scene are near-vertical solar radiation (elevation angle greater than 60°), ambient illuminance greater than 50,000 lux, pupils constrict to their limit diameter, and high-contrast shadows easily form in the eyelid area. This scene will be referred to as [missing information] in the following text. .

[0254] Average gray level of the face when the infrared fill light is off The decision formulas mapped to different environments are as follows:

[0255]

[0256] in It is a constant.

[0257] S2. Before image acquisition, appropriate initial values ​​for optical parameters need to be set according to the environment to prevent images from being severely overexposed or underexposed due to unsuitable initial optical parameters. This would prevent the system from entering the subsequent image quality assessment and feedback adjustment stages because it would be unable to acquire an iris image. Setting appropriate initial values ​​for optical parameters can also accelerate the convergence of the optical adjustment algorithm.

[0258] Step S2 specifically involves adjusting the LED lamp current based on the environmental parameters determined in S1, according to experience. Sensor exposure time Sensor gain Set the corresponding initial values, and the mapping formula is:

[0259]

[0260] in , and It is a constant.

[0261] S3. Acquire several consecutive frames of iris images and input them into the comprehensive quality assessment module to complete the comprehensive iris quality assessment. If a qualified iris image is acquired, the automatic exposure parameter adjustment process is exited and the image is output; if the image quality is unqualified, the image quality feedback adjustment stage is entered; if no iris image is detected, a full-range exposure parameter search is performed.

[0262] S4. Select indicators strongly correlated with brightness (mean grayscale value of the eye region, pupil-iris contrast ratio, and mean grayscale value of the iris) from the results of the comprehensive image quality assessment, and combine them with a statistical model to determine the exposure direction and specific exposure parameter adjustment values. Image exposure parameters are closely related to the infrared LED current and sensor exposure time. Increasing both the LED current and the sensor exposure time can increase image brightness, and the two must be coordinated. Since excessive sensor gain will significantly increase image quantization noise, in the dynamic illumination adjustment strategy of this invention, the sensor gain is a fixed value under the same scene and only changes with scene changes.

[0263] Step S4 specifically involves generating the final exposure parameter decision based on the N consecutive iris images acquired in step S3 using a weighted voting mechanism. Each iris image generates two types of adjustment instructions independently based on the comprehensive quality assessment results: (1) LED light source current adjustment instructions. (2) Image sensor exposure time adjustment command The decision-making rules are as follows: when the quality assessment determines underexposure, output +1 (increase parameters); when overexposure, output -1 (decrease parameters); and when exposure is normal, output 0 (abstain). Voting weights are determined based on the sensitivity grading of exposure-related quality indicators, arranged in descending order of impact: mean grayscale value of the eye region (weight coefficient). Pupil-iris contrast ratio (weighting factor) ), Iris region mean (weighting coefficient) ).

[0264] In the voting process for individual decisions regarding LED light source current adjustment or image sensor exposure time adjustment, each iris image undergoes a sequential voting mechanism based on a preset quality index priority descending sequence (mean grayscale value of eyelid region, pupil-iris contrast, and mean grayscale value of iris). The specific process is as follows: evaluation begins with the highest priority index; the voting for that image is immediately terminated when the current index generates a non-zero vote value (+1 or -1); if the current index evaluation result is an abstention (vote value 0), the process automatically moves to the next priority index for further evaluation. This process continues until a valid vote is generated or all index evaluations are completed, ensuring that each image contributes only one valid vote value to the individual decision.

[0265] For the mean value of the eye region, if its value is less than a certain threshold, it indicates that the image is too dark and the contrast is severely insufficient. In this case, it is necessary to increase the LED current and extend the sensor exposure time to improve the image brightness. If its value is greater than a certain threshold, it indicates that the image is severely overexposed. In this case, it is necessary to reduce the exposure time. Whether to reduce the LED current in this case depends on the environment: if it is an indoor scene, the LED current needs to be reduced; if it is an outdoor scene with strong light or other scenes, the LED current should not be reduced, otherwise it will reduce the light source signal-to-noise ratio and cause the iris texture in the image to disappear. The dynamic illumination adjustment strategy based on the mean value of the eye region is shown in the following formula:

[0266]

[0267] in The mean value for the eye region. and It is a constant.

[0268] For pupil-iris contrast, if the value is below a certain threshold, it indicates that the image is too dark and the contrast is severely insufficient. In this case, it is necessary to increase the LED current and extend the sensor exposure time to improve image brightness. If the value is above a certain threshold, it indicates that the image is severely overexposed. In this case, it is necessary to reduce the exposure time and adjust the LED current according to the environment. The dynamic illumination adjustment strategy based on pupil-iris contrast is shown in the following formula:

[0269]

[0270] in For pupil-iris contrast, and It is a constant.

[0271] If the mean value of the iris region is less than a certain threshold, it indicates that the image is too dark and the contrast is severely insufficient. In this case, it is necessary to increase the LED current and extend the sensor exposure time to improve the image brightness. If the mean value is greater than a certain threshold, it indicates that the image is severely overexposed. In this case, it is necessary to reduce the exposure time and adjust the LED current according to the environment. The dynamic illumination adjustment strategy based on the mean value of the iris region is shown in the following formula:

[0272]

[0273] in This represents the mean value of the iris region. and It is a constant.

[0274] Define the quality confidence of the current image. for:

[0275]

[0276] in, This is the overall quality score for the current image. and These are fixed score thresholds.

[0277] The voting results were weighted and averaged according to the weight of each indicator and the quality confidence level. The final voting result is as follows:

[0278]

[0279] in, The weight of the k-th quality indicator. Let be the quality confidence of the i-th image. By introducing the quality confidence, we can enhance the adjustment effect of high-quality images while suppressing random errors in a single frame.

[0280] After voting is completed, the voting results need to be mapped to specific changes in LED current and exposure time. Let the LED current adjustment step size be... Exposure time adjustment step size is The mapping relationship between the voting results and the changes in LED current and exposure time is shown in the following formula:

[0281]

[0282] in This is the current LED current. This is the current exposure time. For the new LED current, For the new exposure time, and These are the minimum values ​​of LED current and exposure time, respectively.

[0283] S5. Verify environmental consistency. If the environment has not changed, update the sensor exposure parameters based on the result of S4 and proceed to step S3. If the environment has changed, reinitialize the environmental parameters and proceed to step S2. The purpose of environmental consistency detection is to prevent background noise from interfering with environmental detection. In this embodiment, there may be a special use case where the user starts the recognition process first and then faces the iris camera. In this case, the initial environmental detection is easily affected by background noise, leading to misjudgments. Performing environmental consistency detection immediately after quality assessment and optical parameter decision-making ensures that the face is present in the captured image, avoids background noise interference, and enhances the reliability of the optical parameter adjustment process.

[0284] Step S5 specifically involves acquiring an environmental image with the infrared LEDs off and the sensor exposure time and gain set to fixed values. The original facial image is then scaled down using bilinear interpolation, and the grayscale mean is calculated. The current environment is determined based on the grayscale mean. Comparing the result with the previous environmental detection, if the environment remains unchanged, it indicates that the optical parameter adjustments based on the iris image quality are effective. The LED current and exposure time are then updated, and the process jumps to step S3 to begin the next round of feedback adjustment. If the environment has changed, the environmental parameters are updated, and the process jumps to step S2 to reset the initial optical parameter values.

[0285] According to the above adaptive optics parameter adjustment method, the optical parameter adjustment can converge within 3 iterations, which can improve the effective image acquisition success rate of the iris recognition module under complex lighting conditions to more than 90%, which helps to shorten the average recognition waiting time and improve the user experience in extreme environments.

[0286] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of a real-time comprehensive quality assessment method and an adaptive optics parameter adjustment system for outdoor iris recognition scenarios.

[0287] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of a real-time comprehensive quality assessment method and an adaptive optics parameter adjustment system for outdoor iris recognition scenarios.

[0288] An application embodiment of the present invention provides an information data processing terminal, which is used to implement a real-time comprehensive quality assessment method and an adaptive optical parameter adjustment system for outdoor iris recognition scenarios.

[0289] This invention can be implemented in various ways, including hardware, software, and a combination of both. Hardware implementation may employ dedicated logic circuits, while software implementation may be implemented by a microprocessor or dedicated hardware through code stored on a storage medium. Those skilled in the art will recognize that the technology can be implemented by computer-executed instructions or processor control code, which may be stored on various forms of media (such as disks, CD / DVD-ROMs), programmable memories (such as ROM firmware), or provided via data carriers (such as optical or electronic signals). Implementation of this invention may involve the use of hardware circuits, such as very large-scale integrated circuits, gate arrays, logic chips, transistors, etc., or programmable hardware devices, such as field-programmable gate arrays, programmable logic devices, etc., or may employ software executed by different types of processors, or a combination of both, such as through firmware implementation.

[0290] In the verification phase of this invention, the dataset consisted of 304 folders, including 212 folders collected indoors (each containing 15 eye images) and 92 folders collected in both indoor and outdoor environments (each containing 15 indoor and 15 outdoor eye images), for a total of 5940 samples. The samples included a number of low-quality iris images, covering various low-quality types such as low contrast, overexposure, pupil dilation, pupil constriction, and blinking. Before use, the data was manually labeled to determine key information such as pupil parameters and iris outer circle parameters.

[0291] In summary, compared with existing related methods, the method proposed in this invention has comprehensive advantages in terms of speed, accuracy and stability in pupil localization, iris clarity detection, iris segmentation accuracy assessment, eyelash detection in low-contrast iris images and adaptive illumination parameter adjustment. Therefore, it is more conducive to the deployment of iris recognition systems in low-computing-power scenarios.

[0292] I. Specific application areas or related products of this invention.

[0293] This invention provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the real-time comprehensive quality assessment method and adaptive optics parameter adjustment system for outdoor iris recognition scenarios as described in any one of claims 1 to 10.

[0294] This invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor performs the steps of the real-time comprehensive quality assessment method and adaptive optics parameter adjustment system for outdoor iris recognition scenarios.

[0295] This invention provides an information data processing terminal, which is used to implement the adaptive optical parameter adjustment system based on the real-time comprehensive quality assessment method of iris images.

[0296] Five specific embodiments are given below. The first three embodiments are used to fully support the foregoing technical solutions, enabling those skilled in the art to complete the image preprocessing, dual threshold generation, pupil and iris localization, eyelid and eyelash detection, and comprehensive quality scoring, thereby demonstrating the correspondence between the technical means, processing flow, and evaluation results of the present invention. The latter two embodiments are optimized embodiments without deviating from the original technical ideas, to show that the present invention has continuous expansion capabilities and adaptability to complex scenarios.

[0297] Example 1: Under stable, unobstructed outdoor natural light conditions, a single-frame near-infrared iris image is acquired. The original image is first scaled according to a predetermined ratio. Then, starting from the image center, the eye region boundary is determined through horizontal and vertical grayscale projection, with the minimum vertical grayscale integral value used as the vertical reference position of the pupil. Subsequently, a high threshold is used to separate the skin and eye regions. Then, a low threshold is iteratively searched within low grayscale sub-intervals to coarsely segment the pupil and iris. The mean, standard deviation, and medium grayscale entropy of the eye region are statistically analyzed, and the medium grayscale second-order gradient entropy change of the original image and the mean-filtered image are compared to obtain a sharpness index. Then, based on the low threshold results, coarse pupil localization and reliable edge point screening are completed. The pupil circle parameter is obtained using the fitting results, and the outer boundary of the iris is searched based on the joint integration of grayscale intervals. Finally, the upper and lower eyelid curves and eyelash occlusion positions are detected, 18 quality indicators are extracted, and a weighted comprehensive quality score is output.

[0298] Example 2: Image acquisition was performed under outdoor side lighting conditions with localized high-brightness reflections. Since the brightness of the sclera region on one side of the image is significantly increased, a grayscale projection after partition normalization was used to search the eye boundary, avoiding boundary offset caused by directly using global projection. In the dual-threshold generation stage, the high threshold is used to exclude bright background and skin interference, while the low threshold uses dynamic sub-interval search to avoid segmentation failure due to the pupil region being raised overall by reflections. During pupil localization, the direction with the least noise is preferentially selected as the baseline, skipping directions with strong interference from eyelashes and textures. In the iris outer circle search stage, the stability of outer boundary localization under uneven brightness conditions is improved by simultaneously examining the grayscale integrals of the iris and sclera regions. This example demonstrates that this solution can not only be used in ideal environments but also maintain robustness under typical complex outdoor lighting conditions, reflecting the solution's adaptability to outdoor scenarios.

[0299] Example 3: Data acquisition was conducted under conditions where the subject exhibited slight squinting and partial occlusion by the upper eyelashes. The system first extracted reliable candidate eyelid edge points using a vertical gradient map, and then used quadratic curve fitting to obtain the upper and lower eyelid boundaries. For eyelash interference areas, the iris was divided into upper and lower sub-regions, and the cumulative distribution of local standard deviations was calculated. Significant eyelashes were segmented first, and then weak eyelash regions with grayscale close to the background were identified based on topological connectivity. Subsequently, texture richness, effective area ratio, eyelid opening / closing degree, and segmentation accuracy were calculated within the effective iris region. This example highlights that the proposed solution not only has a localization function but also filters and corrects the quality of the effective region when occlusion exists, making the final comprehensive score closer to the actual recognizability. This further demonstrates that the multiple evaluation indicators are not isolated superpositions but rather form a collaborative evaluation mechanism around segmentation quality and recognition usability.

[0300] Example 4: In continuous acquisition mode, multiple frame quality scores are output for the same object, and the results are used to determine the exposure direction. Specifically, several frames of iris images are continuously acquired, and a comprehensive quality score and a brightness-related index are calculated for each frame. If a single frame shows abnormal brightness or momentary occlusion, random errors are suppressed through multi-frame voting, and then the adjustment direction of exposure time and gain is determined. This example is still based on the original comprehensive quality assessment system, but it strengthens the closed-loop relationship between the scoring results and optical parameter feedback at the application level, highlighting the engineering application value of this invention in practical equipment.

[0301] Example 5: In dynamic outdoor scenes where the ambient infrared intensity changes, an environmental image is first acquired, and the current environment category is determined based on the grayscale average. Initial luminous intensity, exposure time, and sensor gain settings are then completed. During subsequent continuous acquisition, if the overall quality score decreases, the exposure decision module first provides an adjustment value, and then the environmental consistency verification step determines whether it is due to a sudden environmental change. If the environment remains unchanged, parameter updates are performed; if the environment changes, environmental initialization is re-executed. This example further strengthens the pre- and post-constraint relationships between environmental determination, image quality assessment, parameter feedback, and consistency verification, giving the overall technical solution a clear closed-loop control characteristic.

[0302] II. Evidence related to the technical effects obtained by the embodiments of the present invention.

[0303] Figure 9 , Figure 10 , Figure 11 Indoor and outdoor iris images acquired using static optical parameters and the adaptive imaging system proposed in this invention are shown respectively. The comparison reveals that under static optical parameters, iris images are prone to contrast degradation due to overexposure or underexposure. By introducing the quality assessment and adaptive adjustment mechanism proposed in this invention, the influence of ambient light variations is effectively overcome, improving the overall image contrast. Therefore, this invention can provide a stable and reliable imaging foundation for outdoor iris recognition systems.

[0304] like Figure 12 The diagram illustrates the distribution of the self-reference iris sharpness evaluation index based on the entropy change of the second-order gradient at medium grayscale, as proposed in this invention, on the collected dataset. For indoor blurred iris images, the relative change of the second-order gradient is small, resulting in a low sharpness index value. For outdoor images, strong light causes an overall decrease in contrast, at which point the normalization effect of the denominator of the sharpness index becomes dominant, leading to an abnormally high index value. If the blurring effect is superimposed, the denominator will further decay, causing the index value to rise further. By setting a reasonable threshold, sharp and blurred images collected in different scenes can be accurately distinguished.

[0305] like Figure 13The image shows a visualization example of the pupil localization algorithm based on reliable edge point fitting proposed in this invention when processing low-quality iris images. The image covers various typical complex interference scenarios, including: half-closed eyes, eyelash interference, low contrast, strabismus, glare from glasses, and pupil constriction caused by strong outdoor light. The visualization results intuitively demonstrate the robustness of the algorithm under the influence of these various low-quality factors.

[0306] like Figure 14 The figure shows the distribution of the iris segmentation accuracy evaluation index based on dynamic contrast proposed in this invention on the collected dataset (taking pupil segmentation accuracy as an example). The initial design purpose of this index is not to evaluate the statistical performance of the algorithm on the dataset, but rather to assess the reliability of localization in a single frame of iris image without reference labels. Essentially a binary classification index, in practical applications, a decision threshold can be set to distinguish the localization results: taking the threshold in the figure as an example, below the threshold indicates localization failure, and above it indicates success.

[0307] like Figure 15 The figure shows visualization examples of the low-contrast eyelash detection method based on multi-feature collaboration proposed in this invention on indoor and outdoor iris images. Typical scenarios selected in the figure include: iris texture interference, indoor low contrast, few eyelashes, half-closed eyes, and outdoor low contrast, etc. The visualization results intuitively demonstrate the eyelash detection capability of this method in different environments.

[0308] like Figure 16 The figure shows the quality score distribution of the iris comprehensive quality assessment method proposed in this invention on the collected dataset. The "long-tail effect" exhibited in the low-score regions directly reflects the low-quality images generated during the acquisition process. Comparing indoor and outdoor data, it is evident that indoor images generally score higher than outdoor images, which aligns with the objective fact that outdoor scene images suffer severe degradation, thus verifying the accuracy of the assessment method. Therefore, this invention can filter images by setting reasonable score thresholds, thereby providing a stable and reliable input basis for iris recognition systems.

[0309] like Figure 17 As shown, typical image examples of iris comprehensive quality scores obtained based on the algorithm of this invention are presented in different intervals.

[0310] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0311] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time comprehensive quality assessment method for outdoor iris recognition scenarios, characterized in that, The method includes: The input raw iris image is scaled. Starting from the image center, the eye region and the vertical reference position of the pupil are determined by combining grayscale projection and local normalization to search the eye boundary; An improved Otsu binarization algorithm is used to generate a high threshold that distinguishes between skin and eyes, and a grayscale sub-interval dynamic search strategy combined with the improved Otsu binarization algorithm is used to generate a low threshold that distinguishes between pupil and iris. Based on the statistical gray mean, gray standard deviation and medium gray information entropy of the eye region, a self-reference sharpness assessment based on the medium gray second gradient entropy change is performed on the eye region. Based on the low-threshold binarization results, the pupil center is coarsely located. The search direction with the least noise is selected to search for reliable pupil edge points from the outside to the inside. The pupil circle parameters are determined by fitting. The grayscale of the iris region is estimated based on the pupil circle parameters, and the outer circle parameters of the iris are located using an improved integral-differential search method based on grayscale range. Detect eyelid curves and eyelash noise areas; Eighteen indicators were extracted, including pupil-iris contrast, sclera-iris contrast, bright-side and dark-side features, iris texture richness, iris radius, effective iris area ratio, pupil-iris radius ratio, iris grayscale mean, eyelid opening and closing, segmentation accuracy, eye strabismus, eye pose singularity, as well as the grayscale mean, grayscale standard deviation, medium grayscale information entropy, and eye clarity. Each indicator was normalized and weighted averaged to output a comprehensive quality score.

2. The real-time comprehensive quality assessment method of claim 1, wherein, The process of searching the eye boundary and determining the vertical reference position of the pupil includes: Calculate the grayscale projection of the scaled image in the horizontal and vertical directions; Using the image center point as a reference, search for the local maximum value of grayscale projection in each partition along the four directions of top, bottom, left, and right; Using the projection value corresponding to the center point as a reference, and combining the local maximum values ​​in each direction, the horizontal and vertical projection curves are partitioned and linearly normalized. Search along four directions from the center point for positions that satisfy the projection gradient jump condition, and use these positions as the eye boundary. The minimum vertical grayscale integral point is calculated within the eye region and used as the vertical reference position of the pupil.

3. The real-time comprehensive quality assessment method as described in claim 1, characterized in that, The process of generating the high and low thresholds includes: Statistical analysis of gray-level histograms of the eye region was performed, and high thresholds for distinguishing between skin and iris were generated using the Otsu binarization algorithm. For grayscale regions below the high threshold, a dynamic interval search strategy is adopted to divide the grayscale range into multiple sub-intervals and iterate them sequentially. The low threshold for distinguishing between the pupil and the iris is determined by the improved Otsu binarization algorithm.

4. The real-time comprehensive quality assessment method as described in claim 1, characterized in that, The pupil positioning and iris outer circle positioning process includes: Based on the low-threshold binarized image, with the vertical reference position of the pupil as the initial point, the mean value of the low grayscale pixel position is iteratively calculated in a predetermined neighborhood to achieve coarse localization of the pupil center. The coarse positioning result is used as the center to radiate the search in multiple directions, and the direction with the least noise is selected as the baseline. Detect pupil boundary points from the outside to the inside along the baseline. The boundary points simultaneously satisfy three conditions: a sudden change in brightness, the gradient of the point in the original image exceeds a predetermined threshold, and the vertical distance from the baseline increases compared to the previous boundary point. The pupil circle parameters were determined by fitting and geometric constraints. The average gray level of the iris is estimated by expanding the annular region based on the pupil circle parameters. Using the pupil center as a reference, the outer circle of the iris is searched within a preset radius. The iris interval and sclera interval are defined based on the low threshold, the average gray value of the iris, and the high threshold. The gray value integral value of the candidate circle in the iris interval and the sclera interval are calculated respectively. The parameters of the outer circle of the iris are determined in a way that maximizes the joint objective.

5. The real-time comprehensive quality assessment method as described in claim 1, characterized in that, The process of obtaining the eyelid curve, eyelash noise, and segmentation accuracy includes: Calculate and binarize the vertical gradient map of the eye region, extract candidate edge points of the upper and lower eyelids, and obtain a quadratic function-type eyelid curve by least squares fitting. The candidate edge points satisfy three constraints: gray level is below the high threshold, the density of neighboring edge points meets the standard, and they are smoothly connected with adjacent points. The iris region is divided into upper and lower sub-regions. The cumulative distribution of local standard deviations at different gray levels is statistically analyzed. The significant eyelash segmentation threshold and the weak eyelash segmentation threshold are determined based on the proportion of the cumulative distribution of the upper sub-region in the complete region. The weak eyelashes are identified by utilizing the topological connectivity between weak and significant eyelashes. The accuracy of pupil segmentation and iris segmentation is evaluated by the dynamic ratio of the sum of the local standard deviations of the segmentation edges to the overall standard deviation of the target region.

6. An adaptive optical parameter adjustment system for implementing the real-time comprehensive quality assessment method for outdoor iris recognition scenarios as described in any one of claims 1 to 5, characterized in that, include: The environment initialization module is used to determine the operating environment based on the ambient infrared intensity. The optical parameter initialization module is used to initialize optical parameters according to the usage environment. The image quality comprehensive evaluation module is used to execute the real-time comprehensive quality evaluation method described in claim 1 to perform a comprehensive quality evaluation of the iris image; The exposure decision module is used to determine the direction of optical parameter adjustment based on the overall image quality, and the optical parameter adjustment value is determined by voting from multiple iris images; The environmental consistency verification module is used to determine whether to perform feedback adjustment according to the exposure decision result based on changes in environmental infrared intensity.

7. The adaptive optics parameter adjustment system as described in claim 6, characterized in that, The environment initialization module is used for: Acquire environmental images with infrared emission turned off; The original face image was scaled down using bilinear interpolation and the mean gray value was calculated. The current environment is determined based on the average gray value.

8. The adaptive optics parameter adjustment system as described in claim 6, characterized in that, The exposure decision module and the environmental consistency verification module are used for: Acquire multiple consecutive frames of iris images and input them into the image quality comprehensive evaluation module; When a qualified iris image is acquired, the automatic adjustment process exits and the image is output. When the image quality is unsatisfactory, the image quality feedback adjustment stage is entered. Perform a full-range optical parameter search when no iris image is detected; Select indicators that are strongly correlated with brightness, and combine them with statistical models to determine the exposure direction and optical parameter adjustment values; To verify environmental consistency, the optical parameters are updated based on the evaluation results when the environment remains unchanged, and the environmental parameters are reinitialized when the environment changes.

9. A method for adjusting the adaptive optics parameters of an iris image based on the real-time comprehensive quality assessment method of claim 1 and the adaptive optics parameter adjustment system of claim 6, characterized in that, The method includes: Acquire environmental images and determine the usage environment based on the environmental infrared intensity and the average grayscale value of the environmental images; Initialize the luminous intensity, exposure time, and sensor gain value according to the described usage environment; Acquire continuous iris images and obtain the comprehensive quality score of each frame using the real-time comprehensive quality assessment method described in claim 1; The direction of optical parameter adjustment is determined based on the overall quality score and brightness-related indicators; The optical parameter adjustment values ​​are determined based on the voting results of multiple frames of images; Parameter updates or environment re-initialization are performed based on the environment consistency verification results.

10. The method for adjusting adaptive optics parameters of iris images as described in claim 9, characterized in that, The optical parameter adjustment based on the overall quality score includes: When the overall quality score reaches the preset requirement, the current iris image is output and the adjustment ends; When the overall quality score does not meet the preset requirements and an iris image is detected, the optical parameters are fine-tuned based on the brightness strong correlation index and statistical model. When no iris image is detected, perform a full-range optical parameter search; After each adjustment, the iris image is reacquired and the quality assessment is repeated until a qualified image is obtained or the environmental reinitialization condition is triggered.

Images