256 results about "Binary segmentation" patented technology

Disclosed is a method of segmenting one or more objects from one or more backgrounds in an image, the method comprising defining a plurality of image nodes, each said image node corresponding to one or more pixels of said image, connecting pairs of adjacent nodes with n-links, each said n-link weighted with an n-link cost, defining a source node, defining a sink node, defining one or more object seeds, said object seeds corresponding to image nodes within said objects, defining one or more background seeds, said background seeds corresponding to image nodes within said backgrounds, connecting said source node with each said object seed with a plurality of t-links, connecting said sink node with each said background seed with a plurality of t-links, wherein each said t-links is weighted with a t-link cost, and calculating a segmentation cut having the smallest total cost of all cuts separating said source from said sink, wherein said total cost of each said cut is defined as the sum of the costs of all said n-links and t-links that each said cut severs.

Method and apparatus for creating foreground masks, or mattes, in images including complex images. A discriminative matting technique may generate accurate alpha mattes for textured images or objects with spatial-varying color distributions. Given an input image and a trimap defining an unknown region, a discriminative color analysis is applied to the unknown region, yielding estimated alpha values, estimated binary segmentation values, and a mixture probability map for the region. The map may be adaptively smoothed. The pixels in the unknown region are classified into boundary pixels and non-boundary pixels according to the probability map. The non-boundary pixels are classified as either foreground or background pixels using a differencing technique that compares multiple pixel features. The estimated alpha values for the boundary pixels are refined. An alpha matte for the image is output. The process may be repeated until convergence of the alpha matte.

Methods and apparatus for adaptive trimap propagation. Methods are described that allow a trimap to be propagated from one frame to the next in a temporally coherent way. A radius-based method propagates automatically computed local trimap radii from frame to frame. A mesh-based method employs pins on the binary segmentation boundary and a mesh generated for the unknown region; the pins are tracked from one frame to the next according to an optical flow technique, the mesh is deformed from one frame to the next according to the movement of the pins, and the adaptive trimap is then warped according to the deformed mesh. These methods can be used separately, or the first method can be used to propagate some regions of the adaptive trimap, and the second method can be used to propagate other regions of the adaptive trimap.

The present invention discloses a road detection algorithm of fusion of area and edge information. The algorithm comprises: collecting road image, and employing self-adaption median filtering and noise reduction to obtain enhanced images; selecting the R component of the enhanced images in the RGB color space, employing a maximum between-cluster variance (Otsu) method to realize the road area segment of the images to obtain a binary segmentation image, and optimizing the binary segmentation image through adoption of the serial mathematical morphology; detecting the binary segmentation image through adoption of the optimal edge detection Canny detector to detect the binary segmentation image and obtain the road edge detection information, and employing the Otsu method to calculate the dual threshold in the optimal edge detection Canny detector; and employing the binary segmentation image and the road edge detection information to obtain a road border line in the image. The road detection algorithm of fusion of area and edge information employs the image processing technology to detect and extract the road information for the autonomous navigation system of a cable tunnel robot; and moreover, the road detection algorithm of fusion of area and edge information is a self-adaption road border line segmentation fusion method and can obtain a smooth and accurate road edge.

The invention discloses a text image super-resolution reconstruction method based on a conditional generative adversarial network. According to the method, the structure of a common conditional generative adversarial network is modified, so that the method is more suitable for a super-resolution reconstruction task of a text image; Non-text binary segmentation image is used as extra training supervision information of a super-resolution reconstruction model, and text-text binary segmentation image is combined with extra training supervision information of a super-resolution reconstruction model; the non-text binary segmentation information constructs a loss function of the model to constrain the training of the model, so that the super-resolution reconstruction model is more concentrated in the text part in the image. Compared with a common image super-resolution method, the text image super-resolution reconstruction method disclosed by the invention fully and specifically utilizes theinformation of the text, and the quality of text image super-resolution reconstruction is effectively improved.

The invention discloses a fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering. The fast high-resolution SAR image ship detection method comprises the following steps: on the basis of the back scattering characteristics of each ground object and the prior information of a ship target in an SAR image, positioning a target potential position index map by an Otsu algorithm and range constraint; on the index map, pre-screening to obtain a detection binary segmentation map by a CFAR (constant false alarm rate) algorithm based on a local contrast; carrying out morphological processing to a detection result, and extracting a potential target slice from the SAR image and a detected binary segmentation map according to a processing result; and carrying out K-means clustering to the extracted slice by a designed identification feature to obtain a final identification result. According to the fast high-resolution SAR image ship detection method based on feature fusion and clustering, the data volume of a detection stage is effectively reduced by pre-processing, and point-to-point detection is not needed / the time of point-to-point detection is saved. Meanwhile, a target identification problem under the condition of insufficient training samples at present can be solved by the designed characteristic and a non-supervision clustering method, the target can be effectively positioned, and the size of the target can be estimated.

The invention discloses a lane line detection algorithm based on an instance segmentation neural network framework. The method comprises the following steps: firstly, establishing an instance segmentation model for lane line detection, wherein the instance segmentation model comprises a binary segmentation module branch and an embedded module branch; loading the original image information into a binary segmentation module branch, and segmenting lane line pixels from background pixels; Loading the original image information into an embedding module branch to obtain a pixel embedding vector of alane line; Then clustering the lane line pixels and the pixel embedding vectors; Transforming the original image by using a transformation matrix network module, outputting a projection transformation matrix of the aerial view, and projecting lane line pixels into the aerial view through the projection transformation matrix; and finally, performing lane line fitting on the pixel points projectedinto the aerial view, and outputting a lane line fitting result. According to the method, the calculated amount is reduced, and meanwhile the influence of environment change on the lane line detectioneffect is eliminated.

A system and method for vertebrae and intervertebral disc localization in magnetic resonance (MR) images is disclosed. In this method, a target spine MR image is segmented by generating a binary segmentation. A spinal cord line is then estimated on the segmented image. Neighboring vertebrae in a region of interest around the estimated spinal cord line are then localized by labeling vertebral regions based on characteristics, such as size and shape, of regions of connected segmented pixels within the regions of interest. A center point of an intervertebral disc between the neighboring vertebrae is then calculated by finding a point halfway between respective borders of the vertebral regions on a line connecting the centers of mass of the vertebral regions.

The invention discloses a fuzzy multi-keyword retrieval method of encrypted data in cloud environment. A file is subjected to set encryption by a data owner to generate a ciphertext file; keywords are extracted from each file; the keywords are subjected to binary segmentation and vectorization to obtain a binary vector group; the binary vector group is subjected to dimensionality reduction and is then inserted into a counting type bloom filter to generate index vectors; each index vector is encrypted to obtain a security index; the ciphertext file and the security indexes are sent to a cloud server; a pre-authorized data user or the data owner extracts the keywords from query data; binary segmentation and vectorization are performed to generate a query vector; encryption is performed to obtain a trap door; the trap door is sent to the cloud server; the cloud server obtains a certain number of files with the highest relevancy degree through query according to the trap door and the security index; after sorting, the files are returned to the data owner. The large data volume of ciphertext multi-keyword retrieval is supported; compared with the prior art, the method has the advantages that the index building and query processes are more efficient; the sorting result is more accurate; the data privacy is effectively protected.

The invention discloses a method for automatic measurement and morphological classification statistic of ultra-fine grain steel grains. The method provided by the invention comprises the following steps of 1, acquiring an image of an ultra-fine grain steel grain and carrying out pretreatment, 2, carrying out binary segmentation of the pre-treated image by a region division-based self-adaptive threshold segmentation method to obtain a binary image, 3, carrying out grain boundary repair of the binary image by a distance transformation-based modified watershed algorithm, and carrying out grain aperture filling by a modified seed filling algorithm to obtain a repaired image, 4, extracting grain morphology characteristic parameters, and 5, carrying out grading statistic of grain sizes according to diameters, and carrying out grain morphology classification according to roundness, form factors and length-width ratios. Through the method provided by the invention, image repair and accurate and efficient measurement, classification and statistic of an ultra-fine grain steel microstructure (grain) can be realized automatically.

The invention discloses an object segmentation method and an object segmentation device based on multilevel local region integration, and calculation equipment, and belongs to the technical field of computer vision. The method comprises the steps as follows: for an image to be processed, selecting multiple local candidate regions according to multiple different preset scales; carrying out image segmentation on each local candidate region and predicting the binary segmentation mask of each local candidate region; carrying out image classification on each local candidate region and predicting the object category to which each local candidate region belongs; and according to the object category to which each local candidate region belongs and the binary segmentation mask of each local candidate region, integrating all the local candidate regions to get an object segmentation image. During detection of an object, the object can be segmented and the boundary thereof can be determined. After the segmentation result of local candidate regions is obtained, an effective way of local region integration is used, and a better optimization result can be obtained.

The utility model discloses a method by use of digital sequence subtraction angiographic images to segment vascular data, which makes use of standard deviation of statistics to describe the changing information of pixel gray value of each point on images along with time. By calculating the gray value standard deviation of each pixel point, 0 to 255 gray space is mapped and a feature graph describing the change of each pixel is obtained. The standard deviation of the area with bigger change of pixel gray value in time, namely, the objective vascular area, is a little bigger. The object is strengthened if reflected in the feature graph. The standard deviation of the area with smaller change of pixel gray value in time, that is the area with no blood vessel, is a little smaller. The background is weakened if reflected in the feature graph; the utility model adopts partial threshold value segmentation method on a basis of busyness measurement, carries out the binary segmentation on the feature graph, then makes use of marking technology of maximum communication domain and the connectivity of vascular tree in structure and wholly segments the vascular tree.

The invention relates to the field of image processing, and discloses a method of extracting a rectangular building from a remote sensing image. The method comprises the steps of obtaining a plurality of superpixel region blocks by superpixel segmentation of a remote sensing image; determining two seed points on a target building in the remote sensing image; merging the plurality of superpixel region blocks based on the determined seed points; carrying out corner detection for the remote sensing image; calculating to generate a corner distance saliency map related to each pixel point in the remote sensing image based on the corner detection result; carrying out binary segmentation for the corner distance saliency map; determining priori information based on the merging result of the plurality of superpixel region blocks and the distance saliency map after binary segmentation; obtaining a building segmentation result by segmenting the remote sensing image based on the priori information; carrying out morphological image processing for the building segmentation result; and obtaining a rectangular target building by rectangular fitting of the building segmentation result after the morphological image processing. The above the method can accurately extract the rectangular building from the remote sensing image.

Provided is an abrasive particle chain self-adaptive segmentation method orienting online ferrographic image automatic identification. Step one, a transmitted light image and a reflected light image respectively provided by a ferrographic sensor are respectively preprocessed and converted into a binary image and a graying image; step two, the reflected light image Imgf is adopted to perform coarse segmentation on the basis of gray scale morphology; step three, fine-multi-scale binary morphological segmentation is performed on the binary image after coarse segmentation; a variable scale corrosion-expansion algorithm is adopted on each abrasive particle chain to realize segmentation of large and small abrasive particles so that a binary segmentation line is acquired; and step four, the binary segmentation line is superposed on the original transmitted light image and the reflected light image so that an abrasive particle image after segmentation can be acquired. An online abrasive particle chain image segmentation problem can be solved, and the method can also be applied to abrasive particle chain automatic segmentation in a conventional offline ferrographic image so that the method is significant for realizing intellectualization and automation of a ferrographic image analysis technology.

The invention relates to a method and a system for identifying the white-leg shrimp disease on the basis of machine vision. The method comprises the following steps of: S1, judging whether an image is an image of a target to be subjected to disease identification, entering the step S2 if judging that the image is the image of the target to be subjected to disease identification, and stopping a program if judging that the image is not the image of the target to be subjected to disease identification; S2, extracting a color feature parameter of the image; S3, carrying out binary segmentation processing on the image; S4, extracting an area feature of the image which is subjected to binary segmentation processing, and calculating the number of pixel points in a target region; S5, carrying out edge detection processing on the image which is subjected to binary segmentation processing to obtain an edge image of the target region, then extracting a perimeter feature of the edge image and obtaining the number of pixels in a target edge region; S6, obtaining a circularity feature parameter by utilizing a ratio of the perimeter to the area of the target region; and S7, obtaining a disease identification result by training the color feature parameter and the circularity feature parameter which are used as training parameters and categorical data sources of a neural network classification algorithm and then classifying the color feature parameter and the circularity feature parameter.

The invention discloses a method for detecting a picture contour based on the combination of a level set and a watershed, which mainly solves the problem that the prior method generates picture over-segmentation results and cannot obtain an accurate target contour. The method is achieved according to the following steps: (1) detecting the target contour of an input picture by using a level set method to obtain an initial target and a background binary segmentation result; (2) using the detected target contour as an internal control mark of the watershed; (3) performing range conversion and watershed conversion on the internal control mark in turn, and using an obtained watershed line result as an external control mark of the watershed; (4) using the internal control mark and the external control mark to correct a gradient picture of the input picture; and (5) performing the watershed conversion on the corrected gradient picture so as to detect the target contour in the picture. The method avoids the picture over-segmentation, has the advantage of positioning the boundary accurately, and can be applied to segmenting foregrounds and backgrounds and detecting target external contours.

The invention discloses a centroid tracking framework based particle filter and mean shift cell tracking method, which mainly solves the problem of low accuracy rate of the traditional cell tracking method. The cell tracking method comprises the following steps of: performing binary segmentation to a video image, and extracting the central position of each cell; tracking the centroid of the cell, and recording the tracking trace of the cell; respectively recording the starting coordinates and the terminating coordinates of the trace into a starting coordinate set and a terminating coordinate set, and selecting a cell to be tracked; further predicting the trace of the cell to be tracked by using particle filter to obtain a predicted coordinate point in the next frame of image; selecting the subsequent tracking trace of the cell to be tracked by using the mean shift method in good time according to the predicted coordinate point; and circulating the steps of prediction and selection till the last frame of image, and completing the tracking of all cells. Compared with other traditional tracking methods, the cell tracking method has improvement in the aspects of tracking effect and accuracy rate and can be used for analyzing motor cells in a medical microscope video image.

The invention relates to a method for extracting iris features based on a pulse couple neural network (PCNN), and belongs to the technical fields of biometric recognition and computer application. The method comprises the following steps of: taking irregular plaques such as recess, furrow, spots and the like in an iris image as intuitive iris features having physical significance; segmenting a pixel set having similar features in the iris image by using pulse couple characteristics of the PCNN or an intersecting cortical model (ICM) and by applying a series of iris enhancement technologies; segmenting out the irregular plaques to obtain a binary segmentation image; and expressing plaques and non-plaques by using 1 and 0, and extracting the iris feature codes.

The invention relates to a three-dimensional image segmentation method based on a three-dimensional improved pulse coupled neural network, which comprises the steps of: expanding an improved pulse coupled neural network (IPCNN) to a three-dimensional plane from a two-dimensional plane; carrying out binary segmentation by using an IPCNN segmentation algorithm principle and an image sequence space relation; smoothing the edge of a region of interest by adopting morphology; and finally, reconstructing the region of interest by using a volume rendering method. The method quickens the operation speed under the premise condition of ensuring the segmentation quality, and achieves more ideal segmentation effect.

The invention discloses an automatic measuring and morphological classification method for particles precipitated from steel, comprising the steps as follows: firstly, the electron micrographs of the target particles precipitated from steel are subjected to image binary segmentation so as to obtain the binary images of the particles; the binary images of the target particles are denoised by a morphological filtering method, a seed filling method is adopted to fill holes, and the particles to be separated are determined by the domain value determined by experience criterion and the separation of agglomerate particles is carried out; the particles after separation are subjected to region labeling; finally, the neural network morphological classification models of the target particles precipitated from steel are established; and results are displayed and output in the form of graph files. The method can obtain ideal measuring and classification effect without omission inspection and re-inspection; the measurement accuracy of particle size can reach plus or minus 2 microns, the particle size distribution anastomotic rate can be more than 91.7 percent, and the anastomotic rate of morphological classification can be more than 90.5 percent; the particle measuring and classification of one view field cost only a few minutes; and the method has excellent universality and can be used in all the particle measuring and classification works with complex backgrounds and morphologies in the material field and biological field.

There is provided a computer-implemented method for detecting a fatty liver, comprising: receiving imaging data of a computed tomography (CT) scan performed using a single source CT Scanner with settings selected for imaging of non-fatty-liver pathology, segmenting a region of the liver by creating a binary image by applying binary segmentation to a sub-set of pixels of the imaging data according to a first set-of-rules, and mapping the region of liver of the binary image to the segmented region of the portion of the liver of the imaging data, calculating liver parameter(s) for the segmented region of the liver from Hounsfield unit (HU) value(s), and detecting the presence of a fatty liver by analyzing the calculated liver parameter(s) according to a second set-of-rules.

The invention discloses a fast image recognition method for a positioning mark point of a mobile robot in an indoor environment, belonging to the computer vision technology. In the method, the angular point of a lampshade on a ceiling can be accurately recognized via binary segmentation, morphological processing, candidate feature point search and classification, and candidate feature point weight centroid evaluation for the acquired image of a camera. The method can not be influenced by the change of the light intensity of the environment, is high in recognition accuracy and high in recognition speed, and meets the requirements for real-time and accurate positioning of the mobile robot.

The invention provides a sub-pixel-level PCB board defect reference point detection method. The method comprises the following steps: reading in a PCB grayscale image 1; preprocessing the PCB image, wherein self-adaptive median filtering denoising is used to remove pepper-and-salt-type noises to obtain a denoised image 2, a method of maximum inter-class variance is used to carry out binary segmentation of the image to obtain a binarized image 3, and a mathematical morphology method is used to process the image to further remove small interference points and holes to obtain an image 4; using acommunicate-domain algorithm to obtain a position and a scale of a preprocessed image feature region; and then enabling reference circle extraction and positioning efficiency and precision to be improved through a sub-pixel edge detection algorithm and fitting of a least square method. Requirements of high-precision reference point extraction and positioning are effectively satisfied, defects of low precision and low speed of previous reference point positioning are overcome, and the method has the characteristics of high applicability and high accuracy.

The invention belongs to the technical field of medical images, and relates to an automatic organ-at-risk sketching method and device based on a three-level convolutional neural network, and a storagemedium. The method comprises the following steps of: preprocessing the three-dimensional medical image, inputting the preprocessed three-dimensional medical image into the first-stage network, the second-stage network and the third-stage network of the trained three-stage convolutional neural network, sequentially identifying the cross section of the organ to be segmented, coarsely positioning the region of interest of the organ to be segmented, and classifying all pixel points in the region of interest; and then outputting a three-dimensional binary segmentation result; carrying out post-processing, edge extraction and edge smoothing on the binary segmentation result to obtain an automatically sketched organ. The three-level cascade convolutional neural network model is formed by cascading three convolutional neural networks, namely a first-level network, a second-level network and a third-level network. The three-level joint neural network has the advantages that priori knowledge isnot needed, the algorithm generalization ability is good, the robustness is high, the speed is high, full automation is achieved, and the segmentation accuracy is high.

The invention discloses a 3D (three dimensional) printing method of transparent pervious concrete samples based on CT scanning. The method comprises the following steps: collecting pervious concrete samples, carrying out CT scanning on the pervious concrete samples to generate two-dimensional slice images, carrying out gray processing on the two-dimensional slice images, characterizing distribution of different colors by using different grey levels, and carrying out filtration and noise reduction treatment; reconstructing a three-dimensional model by using a three-dimensional visualization technology according to the treated two-dimensional images; multiplying according to the number of voxels and the number of images to calculate the size of a required sample model, and cutting the reconstructed three-dimensional model according to the calculated size; and carrying out binary segmentation on the cut model, extracting an aggregate structure, and printing the pervious concrete samples layer by layer by using a 3D printing technology according to an obtained pervious concrete aggregate model. Through the method, the original pervious concrete samples are subjected to CT scanning, 3D modeling and 3D printing technologies and are finally produced into a transparent pervious concrete material; the method provides great help to the research of blockage mechanism and process of the pervious concrete.

The invention discloses an image multi-target joint segmentation method based on multi-tag multi-sample learning, comprising the following steps: first, performing object detection on an input image set to get a candidate object set, and performing segmented object clustering on the candidate object set to get accurate object tags; the, performing salience detection and binary segmentation on the input image set to get a salient region, transferring the object tags in the candidate object set to the salient region, and performing super pixel segmentation on the salient region to get a salient region containing tags; and finally, using an integrated multi-sample multi-tag learning method based on feature random selection to get the object tag of each super pixel in the salient region, thus completing joint segmentation of the input image set.

The invention relates to a method and a device for automatically restoring, measuring and classifying steel dimple images. The device comprises an image acquiring system, an image pretreating part, an image restoring part, an image analyzing part, etc. The image pretreating part is used for performing median filter noise removal and gray level correction on original images acquired by the image acquiring system; the image restoring part is used for performing binary segmentation by using an adaptive fuzzy threshold valve method; boundary deletion and holes in the obtained binary images are processed respectively by using an ultra-erosion and layer-by-layer expansion method and an improved scanning line seed filling algorithm; the image analyzing part is used for performing region calibration on the processed images and setting the dimple diameter as the diameter of the minimum circumcircle of the dimple; and a random dimple region area algorithm is used to measure the dimple area so as to obtain the dimple diameter. After measurement, the measured classification results are output. The invention has the advantages of accuracy, efficiency and convenience, and can be popularized and applied in fracture measurement, analysis and classification with complex backgrounds and shapes in the material filed.

The invention provides a method for identifying a sea ship of an SAR (Synthetic Aperture Radar) image. By use of the method, detection and identification for a sea ship target of an SAR image under different sea clutter backgrounds can be implemented. The method is implemented by the following technical scheme: calculating a global segmentation threshold value of an input sea SAR image based on anaccumulative distribution histogram to obtain an index matrix of a target leakage pixel; for each pixel, self-adaptively selecting an optimal statistical model from a model dictionary through a KS (Kolmogorov-Smirnov) test for clutter modeling, and performing binary segmentation by use of a CFAR (Constant False Alarm Rate) method; removing isolated noise points by use of a local highlight densityon the basis of a binary segmentation result, extracting a connection region, removing the background and calculating corresponding characteristic parameters to obtain a candidate ship target; and constructing a ship target characteristic set by use of a training data set, and calculating the confidence degree of the ship target by a naive bayesian method.

The invention discloses a texture image surface defect detection method based on a depth convolution auto-encoder. The method includes sampling the images and obtaining image blocks to form a trainingset and a verification set, wherein the training set trains an auto-encoder; inputting the image blocks of the verification set into an auto-encoder for processing to obtain a segmentation thresholdreference value; sequentially sampling the images to be detected to obtain image blocks, and inputting the image blocks into the auto-encoder to obtain reconstructed image blocks and feature vectors of the input and reconstructed image blocks; splicing and differentiating the reconstructed image blocks to obtain an initial segmentation image, performing similarity processing on the feature vectorsof each input image block and the corresponding reconstructed image block, splicing and interpolating to obtain an auxiliary segmentation image, multiplying the initial segmentation image and the auxiliary segmentation image element by element, and thresholding to obtain a binary segmentation image. According to the method, the texture surface defect detection model with high universality and robustness is obtained through training under a small number of normal samples, and the defect recognition precision is improved.