134 results about "Local variance" patented technology

The invention provides a face detecting and tracking method and a device. The method comprises the steps of inputting a face image or a face video, preprocessing the face image or the face video in an illumination mode, detecting a face by usage of an Ada Boost algorithm, confirming an initial position of the face, and tracking the face by the usage of a Mean Shift algorithm. According to the face detecting and tracking method and the device, a self-adaptation local contrast enhancement method is provided to enhance image detail information in the period of image preprocessing, in order to increase robustness under different illumination conditions, face front samples under different illumination are added to training samples and accuracy of the face detection is increased by adoption of the Ada Boost algorithm in the period of face detection, in order to overcome the defect that using color of the Mean Shift algorithm is single, grads features and local binary pattern length between perpendiculars (LBP) vein features are integrated by adoption of the Mean Shift tracking algorithm in the period of face tracking, wherein the LBP vein features further considers using LBP local variance for expressing change of image contrast information, and accuracy of the face detection and the face tracking is improved.

An apparatus and image reconstruction method for use in an apparatus including a digital camera are provided. The apparatus and method include driving an image sensor during a short exposure time in an image reconstruction mode, driving the image sensor during a relatively-long exposure time in the image reconstruction mode, and acquiring first and second frame images having different illumination degrees; assigning substantially identical image data of the first and second frame images to local images each having a predetermined size; calculating local variance values of image data of each of the local areas; comparing a local variance value of the first frame image with a local variance value of the second frame images, assigning a high weight to one local image having a high local variance value, and assigning a low weight to the other local image having a low local variance value; applying the determined weights to image data of individual local areas of the first and second frame images, performing image reconstruction, and displaying a reconstructed image.

A method and system detects candidate targets or objects from a viewed scene by simplifying the data, converting the data to gradient magnitude and direction data which is thresholded to simplify the data. Horizontal edges within the data are softened to reduce their masking of adjacent non-horizontal features. One or more target boxes are stepped across the image data and the number of directions of gradient direction data within the box is used to determine the presence of a target. Atmospheric attenuation is compensated. The thresholding used in one embodiment compares the gradient magnitude data to a localized threshold calculated from the local variance of the image gradient magnitude data. Imagery subsets are containing the candidate targets may then be used to detect and identify features and apply a classifier function to screen candidate detections and determine a likely target.

The present invention relates to an image fusion method, especially to an image fusion method based on depth learning. The method comprises: employing a convolution layer to construct basic units based on an automatic encoder; stacking up a plurality of basic units for training to obtain a depth stack neural network, and employing an end-to-end mode to regulate the stack network; employing the stack network to decompose input images, obtaining high-frequency and low-frequency feature mapping pictures of each input image, and employing local variance maximum and region matching degree to merge the high-frequency and low-frequency feature mapping pictures; and putting a high-frequency fusion feature mapping picture and a low-frequency fusion feature mapping picture back to the last layer of the network, and obtaining a final fusion image. The image fusion method based on depth learning can perform adaptive decomposition and reconstruction of images, one high-frequency feature mapping picture and one low-frequency mapping picture are only needed when fusion, the number of the types of filters do not need artificial definition, the number of the layers of decomposition and the number of filtering directions of the images do not need selection, and the dependence of the fusion algorithm on the prior knowledge can be greatly improved.

The invention discloses an automatic scale segmentation parameter selection method for object remote sensing image analysis. Optimum space scale segmentation parameters are determined through calculating an average local variance curve range method; optimum attribute scale segmentation parameter are determined through a local variance histogram estimation method; optimum merging threshold parameters are determined by employing anisotropy space correlation statistics to calculate horizontal and vertical ranges; scale effect evaluation and scale segmentation parameter optimization adjustment on a segmentation result are carried out by employing an index evaluation method. In combination with the geological spatial statistics and a mode identification theory method, the automatic scale segmentation parameter selection method realizes automatic selection of the optimum scale segmentation parameters before segmentation, can clearly explain the determined optimum scale segmentation parameters based on statistics theories, has higher theory credibility and improves information extraction and analysis efficiency and precision for object remote sensing images.

A method for discriminating textured regions and homogeneous or flat regions in an image. A multi-scale or multi-resolution measurement is performed with respect to the image. Alternatively or additionally, the image is analyzed by using a prior measured noise value and a local variance value in or with respect to the image.

The invention discloses an image blind deblurring method based on edge self-adaption. To solve the problems that as for an existing total variation deblurring algorithm, edges and details of images are easily blurred, a de-mean gradient total variation canonical model is built, weighting coefficients are calculated in an iterated mode by means of local variance self-adaption of gradients of the images, and the ability of the deblurring algorithm to restore the edges and the details of the images. The image blind deblurring method comprises the following steps that (1) a blurred image is input, solutions to a gradient-region clear image and a blurring kernel are obtained alternately, and the initial blurring kernel of the blurred image is obtained; (2) the initial blurring kernel is used for conducting primary non-blind deblurring on the blurred image, and an initial clear image is obtained; (3) clustering is conducted on the initial clear image, the mean value and the weighting coefficient in the de-mean canonical model are updated, and a solution to the blurring kernel is obtained again; (4) the new blurring kernel is used for conducting secondary non-blind deblurring so as to obtain a clear image. Experimental results show that the image blind deblurring method based on edge self-adaption has better deburring effect than the prior art and can be used for image restoration.

In a video system, a method and system for mosquito noise reduction are provided. Mosquito noise may be detected by determining a block variance parameter for an image block and a local variance parameter for a portion of the image block. The block variance parameter may be based on serially determined horizontal and vertical variance parameters. A clamping limit is also determined based on the block variance parameter, the local variance parameter, a relative weight parameter, and a mosquito core limit parameter. Pixels covered by the local variance may be filtered and the filtered output may be compared to the original pixel values to determine a difference parameter. Different filter values may be utilized for progressive and interlaced video. A mosquito noise reduction (MNR) difference parameter may be determined based on the difference parameter and the clamping limit and may be utilized to reduce mosquito noise artifacts.

The invention provides an automatic optic inspection method for surface defects of metal cylindrical workpieces and belongs to the automatic optic inspection methods. The automatic optic inspection method is particularly suitable for detecting surface defects of special metal cylindrical workpieces and comprises the following steps: a linear array CCD acquires the unrolled images of a metal cylindrical workpiece, the local binary pattern (LBP) and the local variance (LVAR) are combined, that is, the image variance intensity serves as the weight value to adjust the local texture extraction and measurement result of the LBP, not only the LBP texture space structure mode but also the texture intensity contrast mode is drawn into consideration, finally, the photoelectric image fault features of metal cylinders are accurately extracted, the sizes, the positions, the ranges, the severities and the like of the defects are determined, and the automatic intelligent detection of the workpiece production quality is realized. Therefore, nonuniform lighting caused by the metal materials of the workpieces per se is avoided, the accuracy in detecting tiny defects is improved, the misjudgement rate is lowered, and automatic optic inspection of the metal cylindrical workpieces is achieved.

The invention discloses a convolutional neural network face recognition method and device based on privacy protection, and the method comprises the steps: obtaining a plurality of face images, and carrying out the preprocessing of the plurality of face images; calculating local variances of the plurality of preprocessed face images; constructing a first complex matrix by taking a plurality of faceimages before preprocessing as real part vectors and taking local variances of the plurality of face images after preprocessing as imaginary part vectors; performing double random phase encryption ofa transform domain on the first complex matrix to generate a second complex matrix; and extracting a first component and a second component from the second complex matrix, respectively inputting thefirst component and the second component into the convolutional neural networks for training, connecting the two trained convolutional neural networks to generate a face recognition encryption network, and recognizing the face image through the face recognition encryption network. The method has high safety and can meet different requirements of face recognition.

The invention discloses a local gray level-entropy difference leak detecting algorithm based on an infrared image and belongs to the technical field of detecting. The innovation points of the local gray level-entropy difference leak detection locating algorithm based on an image entropy theory, an algorithm of local entropy of the infrared image to be detected is improved, and a local weighted entropy algorithm, a local variance weighted entropy algorithm, a local gray level-entropy algorithm, a local gray level-weighted entropy algorithm and a local gray level-variance weighted entropy algorithm are achieved. In terms of the six improved algorithms, a large number of experimental tests prove that a detecting capacity of the local gray level-entropy algorithm is the strongest, accordingly target detecting strategies and algorithm processes of the local gray level-entropy algorithm are confirmed, and accuracy and validity of the local gray level-entropy algorithm is verified. The local gray level-entropy difference leak detecting algorithm based on the infrared image not only inherits the advantages of an original local entropy difference method, but also well reflects that one point of the infrared image is presented by entropy information through introduction of gray level information, and the entropy information reflects distribution conditions of gray levels, highlights performances of high gray level areas, sensitively detecting temperature difference among images and acquires leaking targets.

The invention discloses an SAR image noise suppression method based on joint sparse representation and residual fusion and mainly solves the problems of insufficient speckle noise suppression and poor detail keeping of the conventional SAR image noise suppression method. The method disclosed by the invention comprises the following steps: (1), performing block-matching on images to obtain a similar set; (2), performing local variance estimation on the images; (3), performing joint sparse representation on the similar set through local variance and the WSOMP method, so as to obtain a sparse coefficient, and calculating a residual set; (4), performing residual fusion on the residual set, and performing noise reduction through the wavelet soft threshold algorithm to obtain the fused residual; (5), updating a dictionary through the fused residual and the sparse coefficient; (6), performing image reconstruction on the similar block set through the undated dictionary to obtain a de-noised block set; (7), returning the de-noised block set to the original positions of the images, so as to obtain de-noised images. The SAR image noise suppression method obviously improves the SAR image speckle noise reduction effect and can be used for SAR image target recognition and image enhancement.

The invention discloses a self-adaptive reversible digital watermarking method based on integer transformation. The method comprises the steps of watermark embedding and watermark extraction. The step of watermark embedding is double-layer self-adaptive watermark embedding through image block division and block selection. The method comprises the following steps that original images are divided into blocks, the sub-blocks are divided into two parts of a shadow collection and a blank collection, and the shadow collection is embedded firstly; the pixel X of each shadow block is regarded as an embedding unit, the local variance of each shadow block is calculated, the shadow blocks are classified and the smoothness of the shadow blocks is judged; after completion of embedding of the shadow collection, the pixels of the blank collection are predicted by using the embedded pixels so as to complete watermark embedding of a blank spot layer; and embedding of the shadow collection and the blank collection is completed and then the images are merged so that a gray scale image IW after watermark embedding is obtained and the process of watermark embedding is completed. Distortion of the images can be avoided, and the PSNR of the images can be enhanced on the basis of guaranteeing the embedding capacity so that less image distortion under the same embedding capacity can be realized.

The invention provides an automatic hologram reestablishing method. According to the method, at first, a focus evaluation function is designed and can reflect image detail information and the anti-noise property advantage of a variance by utilizing local variance distribution, and the standard deviation of the local variance distribution serves as an index judging the image clarity degree, so that realization of automatic reestablishment becomes possible; secondly, a decision strategy is designed and used for deciding which performance of a focus function value serves as reaction of the target reestablishment distance, and a maximum value of the function value is adopted as a reaction value of the target reestablishment distance; finally, the separation distance is discussed, the separation distance is the step side in search and directly influences the real-time property of automatic reestablishment, a strategy of gradually refining the separation distance from big to small and carrying out feedback in time is adopted for carrying out research, and therefore the operation efficiency is improved. The method can achieve high-accuracy automatic reestablishment and has certain real-time property and universality.

The invention discloses an image processing device and method thereof. The image processing device includes an local variance estimator, a noise detector, a spatial de-noise filter. The local variance estimator estimates each pixel of an input image signal to separately output a local variance value of each pixel, and generates a noise threshold according to the local variance values. The noise detector determines which pixel indicates noise or image according to the noise threshold. The spatial de-noise filter filers the pixel indicating noise to generate an output image signal.

The invention discloses a self-adaptive sampling rate based image sampling method. The method includes the steps: calculating a variance sigma 2 of an input image, sampling the whole image by a first sampling rate f1 when the variance sigma 2 is smaller than a preset first threshold value T1, or sampling the whole image by a sampling rate f3 when the variance sigma 2 is larger than or equal to a preset fourth threshold value T4, or when the variance sigma 2 is larger than or equal to T1 and smaller than T4, making an image layer l equal to 1, partitioning the image into a plurality of image blocks, obtaining a local variance sigma' 2 of each image block, judging whether the local variance sigma' 2 meeting a preset layering stop condition S or not, if yes, calculating a mean M sigma of local variance sigma' 2 of all image blocks, and selecting different sampling rates adaptively according to the mean M sigma in different ranges. Since the sampling rates are dynamically selected according to local features of images, sampling numbers required for image reconstruction can be decreased to different degrees to enable small space occupancy after subsequent image compression and coding. In addition, the self-adaptive sampling rate based image sampling method has the advantages of simplicity and easiness in implementation, wide application range and the like.

A system and a method of generating a template of at least one artifact for use in image correction are disclosed. An image containing the artifact is generated using at least two homogeneous exposures, each generated at a different detector operating temperature. The local variance of grey values at each pixel position in the image is calculated. Each pixel in the image is then classified. A binary image is generated based on the classification. The template is then formed based on both the binary image and the image data containing the artifact.

The invention relates to an angiographic x-ray diagnostic apparatus with an x-ray radiator having an x-ray tube, with a patient positioning table having a radiation-transparent positioning plate and mattress, with an x-ray detector, with an imaging system, an image processing unit and a measuring device to which the output signals of the x-ray detector are fed and which has a first device for determining a value determining the reproduction quality of clinically relevant objects, a second device for determining local variances of the noise in a homogenous image region of different x-ray images and a device for taking the ratio of the output values of the two devices.