Method for extracting multi-layer peak value of image based on FPGA (Field Programmable Gate Array)

Abstract

The invention discloses a method for extracting multi-layer peak values of an image based on an FPGA (Field Programmable Gate Array), which relates to the technical field of image processing and comprises the following steps of: inputting an image, inputting a high-speed image in an FPGA parallel mode, and setting that a plurality of pixels of a row of high-speed image are simultaneously input in each clock period; marking the peak value of the high-speed image, obtaining the complete peak value of the high-speed image by determining the starting point, the peak top and the end point of the peak value of the high-speed image, and marking the peak value of the high-speed image according to the starting point, the peak top and the end point of the peak value of the high-speed image; screening peak values, finding starting point marks for a plurality of pixels input at the same time in the same clock period according to a position sequence, ignoring other marks of subsequent parallel image pixels, only finding peak top marks, only finding peak end point marks after finding the peak top marks, and circularly selecting complete peaks according to a starting point, peak top and end point sequence of the peak values; the method can improve the peak identification accuracy, is small in noise interference degree, carries out parallel processing, and is high in efficiency and real-time performance.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, in particular to a method for extracting multi-layer peak values ​​of an image based on FPGA. Background Art

[0002] Extracting multi-layer peaks of an image, that is, identifying and locating multiple local extreme points or feature points from an image, plays an important role in edge detection, image calibration, 3D reconstruction, and assisting other image processing tasks. Currently, the commonly used peak extraction methods are mainly the following: Sobel edge detector. This method mainly uses discrete differential operators to calculate the approximate gradient of the image. Usually, a Sobel kernel (usually 3x3 in size) is used for convolution operations to obtain the horizontal and vertical gradients of the image, thereby extracting the edge extreme points of the image. Scale-invariant feature transformation. This method constructs a Gaussian pyramid, performs Gaussian difference (DoG) operations in different scale spaces, detects its extreme points and extracts feature points, calculates its direction and scale, and makes it invariant in scale and rotation. Image histogram analysis. This method calculates the grayscale histogram of the image and analyzes the distribution of its grayscale values. Peak detection is performed based on the local maximum in the histogram. Machine learning-based methods. A neural network is trained using a large amount of labeled data to extract image features and peaks. Then, the feature points or peaks in the image are automatically identified and extracted based on the trained model.

[0003] The above four methods for multi-layer peak processing of images all have defects to a certain extent. The Sobel edge detector performs peak detection in a filtering manner, which is extremely sensitive to noise. In the presence of noise, inaccurate peak detection is prone to occur. The scale-invariant feature transformation involves a large number of complex calculations. In practical applications, high-computing equipment is often required to handle tasks that do not require high real-time performance. The image histogram method is also greatly affected by noise, and its calculation structure is simple and cannot process complex images. The machine learning-based method requires a large amount of data for training in the early stage, and requires powerful computing resources for inference. In addition, its model has poor interpretability. In practical applications, especially when the environment changes greatly, the generalization ability of the model may be insufficient.

[0004] Moreover, the above four methods all have the disadvantage of low computing speed and are mostly used on general-purpose processors such as CPU and GPU, making them difficult to apply to the processing of high-frame-rate and high-resolution images. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a method for extracting multi-layer peak values ​​of an image based on FPGA, which can improve the accuracy of peak recognition, reduce the degree of noise interference, parallel processing, and high efficiency and real-time performance.

[0006] To achieve the above object, a first aspect of the present invention provides a method for extracting multi-layer peak values ​​of an image based on FPGA, comprising:

[0007] Input image, input high-speed image in parallel through FPGA, set each clock cycle to input multiple pixels of a row of high-speed image at the same time;

[0008] Marking the high-speed image peak, obtaining the complete high-speed image peak by determining the starting point, peak top, and end point of the high-speed image peak, and marking the high-speed image peak according to the starting point, peak top, and end point of the high-speed image peak;

[0009] Identify the peak value. After finding the starting point mark in the position order of multiple pixels input simultaneously in the same clock cycle, ignore other marks of the image pixels that arrive in parallel, and only find the peak top mark. After finding the peak top mark, only find the peak end mark. Select the complete peak cyclically in the order of the starting point, peak top, and end point of the peak. Determine the number of peaks according to the number of peak tops, and obtain multiple complete peaks.

[0010] Image delay, delaying the input high-speed image by two lines through the first-in first-out queue FIFO;

[0011] The peak point is obtained by weighted square, the effective value of the identified multiple complete peaks is screened to obtain the effective peak, and the weighted square process is performed on the effective peak to obtain multiple groups of peak points of the current row of the image.

[0012] Preferably, marking the high-speed image peak according to the starting point, peak top and end point of the high-speed image peak comprises:

[0013] All pixels arriving at the same clock cycle are judged in parallel, and pixels with values ​​higher than the previous pixel value are marked as the starting point of the peak;

[0014] When the five pixels before and after the current pixel show a trend of first rising and then falling, and the difference is greater than the set threshold, the current pixel is marked as the peak;

[0015] The starting point of the next peak is determined as the end point of the current peak. If the current peak is the last peak, the end point of the image is marked as the end point of the peak.

[0016] Preferably, the multiple complete peaks identified are screened for effective values ​​to obtain effective peaks, including: within the range of the current multiple complete peaks, when the complete peak pixel value is higher than half of the peak value, the current complete peak is deemed valid and the effective peak is obtained, otherwise the pixel point value is reset to zero.

[0017] Preferably, performing weighted square processing on the effective peak values ​​to obtain multiple groups of peak points of the current row of the image includes:

[0018] Multiply the square of the current pixel value in the same complete peak by its column index number, add them up, and use the sum as the dividend;

[0019] The squares of the current pixel values ​​in the same complete peak are accumulated and summed, which is used as the divisor, and the dividend and the divisor are used to obtain the quotient;

[0020] The obtained quotient is used as the peak point of the same complete peak. The multiple complete peaks identified are weighted averaged to obtain multiple groups of peak points of the current row of the image.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention can provide peak recognition accuracy and is less affected by noise. For the determination of the peak value of the input image, the present invention adopts a method for determining the peak in sequence, and its starting point, peak, and end point are determined one by one in sequence to ensure the integrity of the peak without duplication or omission. In addition, the present invention adds a drop constraint condition to the condition of peak trend determination, which reduces the interference of noise on the determination, thereby effectively preventing misjudgment, wrong judgment, and missed judgment phenomena such as judging noise as a peak, merging two peaks into one peak, and missing peaks. For peak determination after the peak is determined, a weighted average method is adopted. The pixels contained in the peak are thresholded and weighted averaged, which can effectively overcome the offset problem of the peak point under noise interference conditions.

[0023] The present invention has high efficiency and real-time performance through parallel processing. FPGA parallel processing is used to process high frame rate and high resolution images. Structurally: multiple tasks such as peak marking, discrimination and weighting algorithms are executed in parallel, and a pipeline is used between multiple task processes, so that the processing processes do not interfere with each other. In data processing: multiple pixel data are received at the same time, and peak marking and peak discrimination are performed on them at the same time. Multiple pixels are processed at the same time, which significantly saves time. Through the parallel processing method of FPGA, the extraction speed of multi-layer peaks of images is greatly improved to meet the needs of high real-time applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1 The first aspect of the present application provides a method for extracting multi-layer peak values ​​of an image based on FPGA, comprising:

[0028] Input image, input high-speed image in parallel through FPGA, set each clock cycle to input multiple pixels of a row of high-speed image at the same time;

[0029] Marking the high-speed image peak, obtaining the complete high-speed image peak by determining the starting point, peak top, and end point of the high-speed image peak, and marking the high-speed image peak according to the starting point, peak top, and end point of the high-speed image peak;

[0030] Furthermore, marking the high-speed image peak according to the starting point, peak top and end point of the high-speed image peak includes:

[0031] All pixels arriving at the same clock cycle are judged in parallel, and pixels with values ​​higher than the previous pixel value are marked as the starting point of the peak;

[0032] When the five pixels before and after the current pixel show a trend of first rising and then falling, and the difference is greater than the set threshold, the current pixel is marked as the peak;

[0033] The starting point of the next peak is determined as the end point of the current peak. If the current peak is the last peak, the end point of the image is marked as the end point of the peak.

[0034] Identify the peak value. After finding the starting point mark in the position order of multiple pixels input simultaneously in the same clock cycle, ignore other marks of the image pixels that arrive in parallel, and only find the peak top mark. After finding the peak top mark, only find the peak end mark. Select the complete peak cyclically in the order of the starting point, peak top, and end point of the peak. Determine the number of peaks according to the number of peak tops, and obtain multiple complete peaks.

[0035] Image delay, delaying the input high-speed image by two lines through the first-in first-out queue FIFO;

[0036] The peak point is obtained by weighted square, the effective value of the identified multiple complete peaks is screened to obtain the effective peak, and the weighted square process is performed on the effective peak to obtain multiple groups of peak points of the current row of the image.

[0037] Furthermore, the multiple complete peaks identified are screened for effective values ​​to obtain effective peaks, including: within the range of the current multiple complete peaks, when the complete peak pixel value is higher than half of the peak value, the current complete peak is deemed valid and the effective peak is obtained, otherwise the pixel point value is reset to zero.

[0038] Specifically, the weighted square processing is performed on the effective peak value to obtain multiple groups of peak points of the current row of the image, including:

[0039] Multiply the square of the current pixel value in the same complete peak by its column index number, add them up, and use the sum as the dividend;

[0040] The squares of the current pixel values ​​in the same complete peak are accumulated and summed, which is used as the divisor, and the dividend and the divisor are used to obtain the quotient;

[0041] The obtained quotient is used as the peak point of the same complete peak. The multiple complete peaks identified are weighted averaged to obtain multiple groups of peak points of the current row of the image.

[0042] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for extracting multi-layer peak values ​​of an image based on FPGA, characterized in that: include: Input image, input high-speed image in parallel through FPGA, set each clock cycle to input multiple pixels of a row of high-speed image at the same time; Marking the high-speed image peak, obtaining the complete high-speed image peak by determining the starting point, peak top, and end point of the high-speed image peak, and marking the high-speed image peak according to the starting point, peak top, and end point of the high-speed image peak; Identify the peak value. After finding the starting point mark in the position order of multiple pixels input simultaneously in the same clock cycle, ignore other marks of the image pixels that arrive in parallel, and only find the peak top mark. After finding the peak top mark, only find the peak end mark. Select the complete peak cyclically in the order of the starting point, peak top, and end point of the peak. Determine the number of peaks according to the number of peak tops, and obtain multiple complete peaks. Image delay, delaying the input high-speed image by two lines through the first-in first-out queue FIFO; The peak point is obtained by weighted square, the effective value of the identified multiple complete peaks is screened to obtain the effective peak, and the weighted square process is performed on the effective peak to obtain multiple groups of peak points of the current row of the image.

2. The method for extracting multi-layer peak values ​​of an image based on FPGA according to claim 1, characterized in that: The step of marking the high-speed image peak according to the starting point, peak top, and end point of the high-speed image peak comprises: All pixels arriving at the same clock cycle are judged in parallel, and pixels with values ​​higher than the previous pixel value are marked as the starting point of the peak; When the five pixels before and after the current pixel show a trend of first rising and then falling, and the difference is greater than the set threshold, the current pixel is marked as the peak; The starting point of the next peak is determined as the end point of the current peak. If the current peak is the last peak, the end point of the image is marked as the end point of the peak.

3. The method for extracting multi-layer peak values ​​of an image based on FPGA according to claim 1, characterized in that: The effective value screening of the identified multiple complete peaks to obtain effective peaks includes: within the range of the current multiple complete peaks, when the complete peak pixel value is higher than half of the peak value, the current complete peak is deemed to be valid and the effective peak is obtained, otherwise the pixel point value is reset to zero.

4. The method for extracting multi-layer peak values ​​of an image based on FPGA according to claim 1, characterized in that: The weighted square processing is performed on the effective peak value to obtain multiple groups of peak points of the current row of the image, including: Multiply the square of the current pixel value in the same complete peak by its column index number, add them up, and use the sum as the dividend; The squares of the current pixel values ​​in the same complete peak are accumulated and summed, which is used as the divisor, and the dividend and the divisor are used to obtain the quotient; The obtained quotient is used as the peak point of the same complete peak. The multiple complete peaks identified are weighted averaged to obtain multiple groups of peak points of the current row of the image.

Images