72 results about "Difference of Gaussians" patented technology

The invention discloses a vehicle detecting method based on Gauss difference multi-scale edge fusion. The method includes the steps that Gauss scale transformation is performed on images to obtain four Gauss images in adjacent scales; according to the four Gauss images in the adjacent scales, the difference operation is performed between the images in the adjacent scales to obtain three Gauss difference images different in scale, edge detection is performed on the obtained three Gauss difference images through a Sobel operator, then edge fusion with the scale upward searching is performed to remove a lot of background edges while edge information of a vehicle is obtained as much as possible, and expansion, closed operation, hole filling and other series of morphological operation are performed on the fused edge images to obtain a connected domain image representing the vehicle; an outside rectangle of the position where the vehicle is located is determined in the original image according to the position information of a connected domain to detect the vehicle. The images in multiple scales are processed, so that algorithm complexity is reduced, operation amount is reduced, efficiency of vehicle detection is effectively improved, and a good detection result is obtained.

An apparatus and method for extracting feature points from an image in a multiprocessor system having a plurality of processors, the method including: dividing an original image into a plurality of regions so as to be allocated to a plurality of processors of the multiprocessor system; performing, by the plurality of processors, blurring operations by levels; dividing the images blurred by levels into a plurality of regions to be allocated to the processors and calculating, by the plurality of processors, differences of Gaussian (DoGs); and generating feature point data according to the calculated DoGs. Because a plurality of processors performs the operations of the method, the total time to extract the feature points from the image is significantly reduced.

The invention relates to a method for image mosaic based on a feature detection operator of second order difference of Gaussian, which carries out mosaic on sequence images which vary in viewpoint, rotation, proportion, illumination and the like to a certain degree. The method provided by the invention adopts zero crossing point detection of second order difference of Gaussian (D<2>oG) pyramid to replace extreme point detection of the original difference of Gaussian (DoG) pyramid so as to extract scale invariant feature points, thereby effectively simplifying the structure of the Gaussian pyramid. The method comprises the steps of: first, extracting image feature points by using an improved SIFT (Scale Invariant Feature Transform) algorithm; then, searching a rough matching point pairs for the extracted feature points through a BBF (Best-Bin-First) algorithm, and purifying the feature point matching pairs by adopting an RANSAC algorithm so as to calculate an invariant transformation matrix H; and finally, completing seamless mosaic for the images by adopting a fading-in-and-out smoothing algorithm. Experimental results show that the method improves the accuracy and the real-time performance of image mosaic, can well solve problems such as illumination, rotation, scale variation, affine and the like, and realizes automatic mosaic without manual intervention.

The invention pertains to the technical field of image processing and relates to a feature extraction and description method based on SIFT and with invariance in mirror inversion. The method comprises the following steps: (1) a Gaussian kernel convolution process is conducted to an input image; (2) a Gaussian difference process is conducted to the image to detect an extremum point of the image; (3) feature points are selected; (4) locations of the feature points are accurately positioned; (5) a direction parameter of each feature point is determined; (6) summation is conducted to values of gradient modulus respectively taking a main direction as a boundary, at the two sides; and (7) pixel units in a Gaussian weight window are organized to conduct encoding and normalization and further generate description data of the image. The method increases the robustness of feature extraction and description methods to a mirror image and extends the applicable field of computer vision.

A hardware architecture is applied to the calculation of a Difference-of-Gaussian filter, which is typically employed in image processing algorithms. The architecture has a modular structure to easily allow the matching of the desired delay / area ratio as well as a high computational accuracy. A new solution is provided for the implementation of multiply-accumulators which allows a significant reduction of area with respect to the conventional architectures.

The invention aims to provide a fingerprint enhancement method. The fingerprint enhancement method includes the following steps that A, a fingerprint image is normalized to obtain a normalized fingerprint image, and the continuous directional diagram of the fingerprint image is calculated; B, a fingerprint enhancement image is obtained after the normalized fingerprint image is subjected to Gabor filtration, and an initial filtering image is obtained by recombining a continuous directional diagram; C, Gaussian blur is performed on the initial filtering image to obtain a Gaussian blur image, and Gaussian difference is performed on the Gaussian blur image to obtain a Gaussian difference image; D, after being mapped in the opposite direction, the Gaussian difference image is reconstructed to obtain a final filtering image. The invention further provides a fingerprint recognition device. According to the fingerprint enhancement algorithm, the Gabor filtering and the Gaussian difference are combined, and thus the denoising capacity is enhanced, and the defects that an existing fingerprint enhancement algorithm is large in calculation amount and poor in enhancement effect are overcome. Through the fingerprint enhancement method, the fingerprint recognition device is combined with the processing structure of a DSP and the processing structure of an FPGA, and thus the fingerprint recognition efficiency and the fingerprint storage efficiency are improved.

The invention discloses a method. The method comprises the following steps: (a) inputting a two-dimensional grayscale image (x, y); (b) based on the two-dimensional grayscale image and a scale factor, creating a difference of Gaussian scale space, wherein the number of layers of the difference of Gaussian scale space is dependent on the scale factor; (c) selecting one or more remarkable regions from the layers in the scale space; (d) carrying out region description on the selected remarkable regions by using region descriptors; and (e) respectively taking a reference image and an image to be matched as the two-dimensional grayscale image for obtaining the remarkable regions and the related region descriptors of the two-dimensional grayscale image and the reference image, and, based on the Euclidean distances of the obtained region descriptors of the two-dimensional grayscale image and the reference image, carrying out region matching on the remarkable regions of the two-dimensional grayscale image and the reference image.

The invention relates to an image reconstruction method, which comprises the following steps: shooting a group of two-dimensional images around a foot or shooting a complete video around the foot by using a mobile shooting device; performing grayscale processing on the uploaded image; determining feature points in each of the images varying in resolution by forming an image resolution pyramid; performing Gaussian blur at different scales to the images, and sampling the smooth images in a downsampling mode which means that the resolution decreases layer by layer; building a Gaussian pyramid anda Gaussian difference pyramid, wherein in each layer of the pyramids, the position representing the Harris angle feature of the images is determined; finding the position of each corner feature, displaying the angular strength, according to the comparison between the angular strength and the predetermined minimum strength threshold, deleting the points, repeating the tests, determine whether thetotal exceeds the maximum permitted value, if so, increasing the radius again and repeating the process, and alternately performing the operations until the number of corner positions equals or decreases to the maximum permitted value.

The invention belongs to the technical field of image processing and relates to a method for extracting and describing image characteristics with turnover invariance. The method comprises the following steps of: (1) carrying out Gaussian kernel convolution processing on the input image; (2) carrying out Gaussian processing on the image continuously and measuring the extreme point; (3) screening feature points; (4) precisely positioning the position of the feature point; (5) determining the direction parameter of each feature point; (6) summing the gradient magnitudes taking the principal directions as the boundary line at two sides respectively; and (7) dividing auxiliary regions, organizing pixel units and carrying out coding and normalized operation, thus generating the describing data on the image. The method increases the haleness of feature extraction and description method on the mirror surface imaging problem and enlarges the application field of computer vision.

The invention provides a face recognition method and device, and relates to the technical fields of pattern recognition and biological feature recognition. The face recognition method and device can be used for carrying out effective recognition under various illumination conditions, and enhances the performance of face recognition. The method comprises the following steps: acquiring an original face test sample to be recognized, and carrying out Gaussian difference filter treatment on the original face test sample to obtain a filtered face test sample; comparing the original face test samplewith a prestored original face training sample set registered under normal illumination conditions, comparing the filtered face test sample with a prestored face training sample set subjected to Gaussian difference filter, and finding out the recognition object corresponding to the original face test sample from the original face training sample set; and computing to obtain an overall SCI (Sparsity Concentration Index) of the original face training sample of the recognition object and the filtered face training sample, and judging whether the original face test sample is a registered face according to the SCI.

The invention discloses a hardware circuit implementation method for an image matching descriptor generation scale space. The method comprises a step of loading an image to be matched in a specific application, a step of performing Gaussian smoothing on the image to generate a first group of images, a step of generating a next group of images by using a method of the combination of downsampling and Gaussian filtering, and a step of generating a Difference of Gaussian pyramid according to an obtained Gaussian pyramid. By redesigning a scale space generation method of an SIFT image matching algorithm to adapt to the characteristics of fixed point, parallel and pipeline of an FPGA operation, the speed of scale space generation is accelerated, hardware resources are saved, and therefore, the performance of a whole feature matching module is effectively optimized.

The invention, which belongs to the field of digital image processing, discloses a Har-SURF-RAN feature point matching method for stereoscopic visual three-dimensional reconstruction. The method comprises: step one, Gaussian difference preprocessing is carried out on pixel points in an image, candidate feature points are screened, and screening is carried out again by using a weight coefficient toobtain feature points; step two, a feature point data set is processed to obtain feature descriptors of the feature points; step three, the detected feature points are matched by using nearest neighbor feature matching; and step four, false determination points in feature matching point pairs are rejected preliminarily, a data set is sorted by using the idea of PROSAC, an RANSAC selection model and evaluation parts are improved, the RANSAC is improved by using the local optimization idea, and the local part of the model is optimized further by introducing an optimization parameter K. Therefore, the computing load is reduced; the speed is increased; redundant points generated by the improved algorithm are reduced; and the precision is improved.

The invention relates to the technical field of display screens, and provides a defect detection method and device for an AMOLED display screen, computer equipment and a storage medium. The defect detection method for the AMOLED display screen comprises the steps: acquiring a to-be-detected image containing a target AMOLED display screen; extracting a target area image in the to-be-detected image;preprocessing the target area image to obtain a residual image; calculating a local maximum value point of the residual image through a Gaussian difference algorithm, and taking the local maximum value point as a seed point; and obtaining the defect position of the target AMOLED display screen according to the seed point and a preset threshold. Through the implementation of the defect detection method, the problem that in the prior art, the detection process of an AMOLED display screen is complex can be solved.

The invention discloses a license plate positioning algorithm fusing an affine invariant corner feature and a visual color feature. A corner with a scale and affine invariant feature and a color feature are extracted in a Gaussian difference scale space. A rapid fusion method based on a multi-scale product is provided in multi-scale corner feature and multi-feature fusion. A license plate is accurately positioned through distance and intensive relationship between license plate region feature points. The test of a lot of vehicle images actually shot in a complex environment shows that the license plate positioning algorithm is rapid and efficient, and has a great robust performance in terms of rotational transforming, scaling and noise.

The invention belongs to the technical filed of computer vision and especially relates to brightness enhancement and edge enhancement of a grayscale image. The method specifically comprises the following steps: estimating global brightness and determining algorithm self-adaption parameters, generating a brightness mapping graph of the images and carrying out calculation to obtain a brightness enhancement image and carrying out edge enhancement processing. The method is characterized by, to begin with, estimating the self-adaption parameters according to the brightness distribution conditions of global dark areas; then, carrying out global brightness enhancement processing on the images and obtaining the modulation mapping graph of the whole picture through a modulation function, and carrying out calculation to obtain brightness enhancement result; and finally, realizing edge enhancement based on a dimension-self-adaptive Gaussian difference model, the model dimension being influenced by contrast ratio, and therefore, bright areas can be enhanced with finer texture information, and the dark areas can be enhanced with larger contour information. The method can enhance overall brightness and contrast of the grayscale images effectively, and the self-adaption characteristic can play a very good effect on edge enhancement of the bright and dark areas.

The invention discloses a calligraphy inverse hanging prevention method based on SIFT (Scale invariant feature Transform). The method comprises the following steps that: firstly, preprocessing a calligraphy font image to obtain the binary image of a calligraphy font, carrying out convolution operation, and importing a DOG (Difference of Gaussian) scale space operator to generate a DOG scale space; then, detecting the extreme value point of the calligraphy font image; and finally, carrying out operation to obtain a Sift feature descriptor, and matching the Sift feature descriptor with radicals in an existing radical library to obtain a proximate result. By use of the method, the problem in the prior art that foreign friends inversely put Chinese characters since the foreign friends do not understand Chinese characters can be solved.

The invention discloses a steel rail surface defect detection method based on reverse Gaussian difference. The method comprises the following steps that S1, acquiring a steel rail surface panoramic image through an image acquisition device; s2, extracting a steel rail image of a target area part from the steel rail surface panoramic image obtained in the step S1 by using a vertical projection method; s3, performing reverse Gaussian filtering on the steel rail image extracted in the step S2 to obtain a steel rail reverse Gaussian filtering image; s4, performing difference on the steel rail image extracted in the step S2 and the steel rail reverse Gaussian filtering image obtained in the step S3 to obtain a steel rail difference image; s5, binarizing the steel rail difference image in the step S4 to obtain a steel rail binarized image; and S6, carrying out area filtering and closing operation on the steel rail binary image in the step S5 so as to complete the detection of the defect areaon the surface of the steel rail. The method is suitable for various different rail environments, and a very good steel rail surface defect detection effect can be obtained.

The invention discloses a free liquid level identification and extraction method of a watershed algorithm based on Gaussian filtering, belongs to the technical field of image processing, and is used for solving the technical problem of difficulty in identifying and extracting a free liquid level of an existing liquid level image. The method comprises the following steps: (1) collecting a liquid level picture, and converting the picture into a binary image through gray processing; and (2) carrying out convolution processing on the image after gray processing by using Gaussian difference filtering, highlighting the edge with the large gradient in the image gray value, weakening the noise, weakening the edge with the small gradient, and achieving the purpose of highlighting the edge of the free liquid level; (3) performing watershed segmentation on a gradient image formed after Gaussian differential filtering, and identifying and extracting a free liquid level; and (4) superposing the extracted free liquid level back to the liquid level image, and checking the segmentation effect. The method can accurately identify and extract the free liquid level of the image, and can be applied tothe technical field of physical model experiments of storage cabin carrying equipment in aerospace, transportation and the like.

The invention provides an SAR image and optical image matching method based on feature matching and position matching. The SAR image and optical image matching method comprises the following steps: performing preliminary key point detection on an optical image and an SAR image by using a Gaussian difference algorithm; according to the detected key points of the optical image and the SAR image, extracting a surrounding area, and reconstructing an image block; designing a deep convolutional neural network comprising a dense block and a transition layer, designing a composite loss function, and generating a deep feature descriptor by training and operating the deep convolutional neural network; performing feature matching on the optical image and the SAR image by using an L2 distance algorithm and a depth feature descriptor, and evaluating a distance error of a matching point; and realizing position matching of the SAR image and the optical image through a two-dimensional Gaussian function voting algorithm. The problem of feature matching of the SAR image and the optical image is solved, better matching capability and accuracy are achieved, and position matching of the SAR image and the optical image can be achieved.

The invention provides a shoeprint wear area detection and stroking method. The shoeprint wear area detection and stroking method comprises: preprocessing read-in image features to obtain an area of interest of a to-be-processed image; detecting a specific point set of the preprocessed image by using a multi-scale Gaussian difference operator; performing conditional screening on the specific pointset to obtain a candidate point set; preliminarily drawing a worn mask by adopting a bilateral region growing method based on gray scale and space distance; removing solid small patterns by adoptinga connected domain-based condition screening method; and performing wear degree grading according to the entropy of the preprocessed image area corresponding to each connected area, and performing labeling of different colors according to grading boundaries. According to the method, the worn part of the sole pattern can be effectively detected, the boundary can be accurately described, the methodhas good adaptability to different types of sole patterns, in the suspect footprint analysis and recognition process, workers are better assisted in shoeprint comparison and analysis more quickly, andthe effect more reliable than that of manual measurement is achieved.

The invention discloses a rock foundation pit blasting parameter dynamic regulation and control method. The method comprises steps of establishing a nonlinear mapping relationship between an input vector and an output vector by using a GP machine learning method; taking the optimal hyper-parameter in the GP as a population sample in the DE algorithm, searching each optimal hyper-parameter of the GP through variation, intersection and selection in the DE process by means of the optimization capability of DE, optimizing the GP model, and improving the prediction capability of the GP model, thereby predicting an output value closer to an optimal solution. Furthermore, the optimal output value of the prediction result of the GP model and the mean square error between the output value and the control value are used as the fitness function of DE, and the prediction capability of the GP model is improved by optimizing the sample population and gradually approaching the optimal solution, so that the target vector closer to the optimal solution is predicted. According to the method, a Gaussian-differential evolution algorithm (GP-DE) blasting intelligent optimization algorithm is introducedon the basis of blasting monitoring data analysis to quickly and effectively optimize the blasting parameters, so that the construction process is ensured to be carried out safely.

The invention provides a fruit flaw classification method and device based on machine vision and deep learning fusion, a storage medium and computer equipment. The method comprises the following steps: acquiring a color image of a fruit by using a camera, and respectively performing background segmentation algorithm processing, background region removal, HSI color transformation, Gaussian difference operation of an S space and the like on the acquired color image, thus obtaining a DoG image; then carrying out threshold segmentation on the DoG image, obtaining a defect region, positioning a target region in a color image, intercepting an image of the defect region, carrying out processing classification, endowing different label numbers, and constructing and training a differential convolutional neural network structure; and obtaining a network connection weight matrix, thereby completing defect classification of the to-be-detected image. According to the method, fruit classification is achieved; the advantages of machine vision and deep learning are fused, and the complexity of fruit flaw classification and recognition is fully considered, so that the recognition rate is improved, meanwhile, the recognition time is shortened, and the interference of fruit stems and calyx on classification and recognition due to angle and posture transformation is reduced.

The invention discloses an intelligent decision support system design and implementation method for unmanned aerial vehicle SAR image automatic target recognition. The method comprises constructing a feature library and an inference engine. The specific process of constructing the feature library is as follows: 11) performing segmentation detection on a target in the SAR image; 12) carrying out target false alarm elimination and target calibration on the SAR image; 13) constructing a multi-scale space based on a Gaussian difference model; and 14) taking multi-resolution features of the SAR image, and constructing a feature library according to the extracted multi-resolution features. The specific process of constructing the inference engine is as follows: 21) constructing a training set and a test set by using the extracted multi-resolution features of the SAR image; 22) constructing a multi-scale multi-core support vector classifier; and 23) training and testing the multi-scale multi-core support vector classifier, and performing automatic target identification of the SAR image by using the trained and tested multi-scale multi-core support vector classifier to construct an inference engine. The method can realize automatic target identification of the SAR image.

The invention discloses an eyeground image blood vessel segmentation method based on self-adaption difference of Gaussians. The segmentation method comprises steps of 1) extracting colorful eyeground image green channels, and performing pretreatment of self-adaption histogram equalization and anisotropy coupling diffusion with limited contrast ratio; 2) constructing Gaussian scale space; 3) subtracting adjacent two layers in the Gaussians scale space so as to get difference of Gaussian images; 4) averaging difference of Gaussian image weighing so as to get blood vessel increasing images; 5) performing binaryzation for the blood vessel increasing images; 6) rotating Gaussian kernel in 12 directions, wherein 15 DEG is regarded as step size in 0-180 DEG, repeating steps of 2-5, and overlapping results in the 12 directions; 7) selecting 20% of the grey value of the second peak as a light area of a threshold value extracting images according to bimodality of a pretreatment image histogram; and 8) reducing light areas from the blood vessel binary images so as to reduce effects of the light areas on segmentation of blood vessels. The method is widely applicable for blood vessel segmenetaion of all kinds of colorful eyeground images.

The disclosure discloses a reconfigurable hardware acceleration method and system for Gaussian pyramid construction and belongs to the field of hardware accelerator design. The system provided by the disclosure includes a static random access memory (SRAM) bank, a first in first out (FIFO) group, a switch network, a shift register array, an adder tree module, a demultiplexer, a reconfigurable PE array, and a Gaussian difference module. In the disclosure, according to the requirements of different scenarios and different tasks for the system, reconfigurable PE array resources can be configured to realize convolution calculations of different scales. The disclosure includes methods of fast and slow dual clock domain design, dynamic edge padding design, and input image partial sum reusing design.

The invention provides a multi-scale face detection method based on angular point skin color detection. The method mainly comprises the following steps: (1) an image angular point extraction method based on Gaussian difference and edge gradient filtering; (2) a skin color corner filtering method; (3) a multi-scale face detection method based on density clustering. The center and the contour edge of the human face are determined by detecting the positions of the skin color angular points in the human face, and therefore, human face detection is achieved.

The invention provides an image edge detection method and device, electronic equipment and a storage medium. The image edge detection method comprises the following steps: for a current pixel in a to-be-processed image, determining an adjustment parameter corresponding to the current pixel according to a pixel value of the current pixel; for each neighborhood pixel of the current pixel, determining a first spatial domain weight and a second spatial domain weight of the current neighborhood pixel according to a coordinate difference value between the current neighborhood pixel and the current pixel; determining a Gaussian difference filtering parameter corresponding to the current pixel according to the adjustment parameter, the pixel value of each neighborhood pixel of the current pixel, the first spatial domain weights and the second spatial domain weights; and determining an edge characteristic value of the current pixel according to the Gaussian difference filtering parameter corresponding to the current pixel. Edge detection is more accurate, and the detection effect is better.

The invention discloses a method for calculating a ventilation resistance coefficient of a laneway based on image recognition, and belongs to the field of mine intelligent ventilation. The method comprises the following steps: photographing surrounding rocks of a wind resistance roadway to be measured to obtain a plurality of pictures containing roughness information, identifying the roughness in the pictures by adopting a Gaussian difference algorithm, and extracting key points of the roughness in the pictures; converting two-dimensional information into a world coordinate system by combining internal and external parameters of the camera and coordinate system transformation, and constructing a point cloud three-dimensional model; segmenting the point cloud model into several areas, fitting a three-dimensional plane through scattered points in each area, and calculating the mean value of the distance from each point cloud to the plane as the roughness of the area; calculating the roughness of other segmented regions by the same method; calculating the average value of the roughness of all the segmented areas to serve as the roughness of the photo; identifying the roughness of the surrounding rock in other photos by the same method, and calculating the average value of the roughness identified by all the photos as the roughness of the roadway to be measured; and finally, combining a Colebrook mathematical model and a friction-resistance mathematical model to predict a resistance coefficient. The method has the advantages of being high in intelligent degree, high in working environment adaptability, capable of achieving data operation and feedback and capable of providing technical support for mine intelligent ventilation.

The invention discloses an infrared weak and small target detection method based on local contrast and gradient, and relates to the field of infrared small target detection, and the method comprises the steps: 1, constructing a Gaussian difference pyramid, and extracting an interest point of an image; step 2, eliminating edge response through adaptive threshold segmentation; step 3, constructing a multi-scale space to carry out extreme point detection; 4, enhancing the target by using a local contrast calculation method; and step 5, obtaining a detection result through gradient direction calculation. By using the detection method of the invention to detect the infrared weak and small target in various scenes, the target detection rate is up to 0.989, the target detection rate is high, and the robustness is better. According to the invention, extreme point detection is carried out by using the Gaussian difference pyramid, so that the extreme point detection precision and effect are improved. According to the method, local contrast calculation and gradient calculation are combined, background clutters are effectively suppressed, the target detection rate is improved, the false alarm rate is reduced, and infrared weak and small targets are detected in various scenes.

The invention discloses an SIFT-based image registration method, device and system and a storage medium, and the method comprises the steps: obtaining a to-be-registered image and a reference image, carrying out the preprocessing of the to-be-registered image, and constructing a Gaussian difference scale space; extreme points in different scale spaces in the Gaussian difference scale space are obtained, feature points are selected according to the extreme points, and feature descriptors are generated for the feature points; calculating a local information entropy of the feature point field, and determining a final feature descriptor of the feature point according to the local information entropy and the feature descriptor; and optimizing feature point matching according to a genetic algorithm, generating a matching point pair, and correcting the matching point pair to complete image registration of the to-be-registered image. According to the method, the number of feature points needing to be matched and the areas needing to be matched are reduced, the matching efficiency and the algorithm speed are optimized, and the search efficiency and the registration precision of the SIFT algorithm are improved.