261 results about "Over segmentation" patented technology

A system and associated method automatically discover salient segments in a speech transcript and focus on the segmentation of an audio/video source into topically cohesive segments based on Automatic Speech Recognition (ASR) transcriptions. The word n-grams are extracted from the speech transcript using a three-phase segmentation algorithm based on the following sequence or combination of boundary-based and content-based methods: a boundary-based method; a rate of arrival of feature method; and a content-based method. In the first two segmentation passes, the temporal proximity and the rate of arrival of features are analyzed to compute an initial segmentation. In the third segmentation pass, changes in the set of content-bearing words used by adjacent segments are detected, to validate the initial segments for merging them, to prevent over-segmentation.

The invention discloses a ground point cloud segmentation method based on a three-dimensional laser radar. The ground point cloud segmentation method comprises the steps of: 1) acquiring polar coordinate data of three-dimensional laser radar scanning point cloud in the surrounding environment of a vehicle and converting the polar coordinate data to a local rectangular coordinate system; 2) correcting radar data by utilizing a vehicle-mounted IMU and an odometer; 3) constructing a polar coordinate grid map, and extracting an extension vertex in each grid according to vertical continuity of point cloud distribution in the grids; 4) extracting ground points in non-marginal grids according to height attributes of the extension vertexes and a ground smooth consistency criterion, and adopting a 3sigma criterion for further extraction of ground points in marginal grids. Compared with the prior art, the ground point cloud segmentation method can reduce ground extraction errors caused by motion of the vehicle, avoids the occurrence of over-segmentation and under-segmentation, is high in precision and high in reliability, has high real-time performance and can be widely used in the field of radar-based environment sensing technologies.

The invention provides an image segmentation method based on a watershed algorithm and a morphological marker. The method comprises the steps that median filtering is carried out on a gray level image to obtain a filtered image; an OTSU method is carried out on the filtered image to obtain a binary image; the binary image is processed through a morphological algorithm based on reconstruction to obtain a characteristic marked image; the characteristic marked image is transformed through the watershed algorithm to obtain a segmented image. According to the image segmentation method, the OTSU method and median filtering are utilized for filtering out impurities and noisy points in the image, the image is adopted as the primary mark source of the watershed algorithm, and the interference of noise is effectively eliminated; a morphological operation method is adopted, the information of an effective area cannot be lost, meanwhile, certain fuzzy areas or connected areas can be separated, and the integrity and consistency of image segmentation are guaranteed. Connected domain calculation is combined, the invalid target and information of non-noisy points can be removed, the marker of the watershed algorithm is precisely located, and the over-segmentation phenomenon is eliminated.

The invention discloses a color image segmentation method, which comprises the following steps: processing an image to be segmented into a gray image; extracting a region profile map from the gray image by an Otsu threshold segmentation method, and determining an optimal threshold; taking the optimal threshold as a high threshold of a Canny operator, and extracting an edge map from the gray image by utilizing the Canny operator; and fusing the area profile map and the edge map, and outputting a segmentation result of a color image. The method adopts an operator threshold determining scheme combined with an adaptive strategy and an empirical value and overcomes the defect that an empirical value selecting method has application limitation; the local change intensity of a gray value is obtained through edge detection, so that over-segmentation of avoidable areas is limited; and residual edges are complemented through region segmentation, so that profiles of images after the segmentation are more clear and complete.

The invention provides an airborne laser point cloud classification method based on a high-order conditional random field. The airborne laser point cloud classification method specifically comprises the following steps: (1) point cloud segmentation based on DBSCAN clustering; (2) point cloud over-segmentation based on the K-means cluster; (3) construction of a point set adjacency relation based onthe Meanshift clustering; and (4) construction of a point cloud classification method of a high-order conditional random field based on the multi-level point set. The method has the advantages that:(1) a multi-layer clustering point set structure construction method is provided, and a connection relation between point sets is constructed by introducing a Meanshift point set cluster constrained by category labels, so that the categories of the point sets can be classified more accurately; (2) a multi-level point set of the non-linear point cloud number can be adaptively constructed, and information such as the structure and the shape of a point cloud target can be more completely represented; and (3) a CRF model is constructed by taking the point set as a first-order item, and higher efficiency and a classification effect are achieved, so that a higher framework is integrated, and a better effect is obtained.

The invention belongs to the technical field of laser radars, and discloses a depth map-based three-dimensional laser radar point cloud target segmentation method, which comprises the following stepsof: converting three-dimensional point cloud data acquired by a three-dimensional laser radar into a two-dimensional depth map; calculating an angle value formed by two adjacent points in each columnin the depth map, and traversing to obtain an angle matrix corresponding to the depth map; traversing the depth map through a breadth-first search algorithm, if the angle difference between two pointson adjacent positions of the depth map is smaller than a specified threshold value, marking the depth map as the same type, thereby finding out the part, belonging to the ground, in the depth map, and removing ground point cloud data according to the mapping relation between the point cloud and the depth map; and carrying out target segmentation on the non-ground point cloud based on an improvedDBSCAN algorithm, and judging whether the point cloud is a core point or not according to an adaptive parameter eps while considering a spatial Euclidean distance and an angle distance. According to the method, the segmentation efficiency on the depth map is improved, the real-time requirement is met, and the problems of under-segmentation and over-segmentation are effectively solved.

The invention discloses an image segmentation method fusing color and depth information. According to the method, firstly, a meanshift algorithm is used for segmenting an input color image to obtain an over-segmentation region set, and then similarities among all the regions are calculated and include color similarities, depth similarities and fusions of the color similarities and the depth similarities; then according to a depth image, seed regions of a target and seed regions of a background are automatically selected; finally, an MSRM algorithm is used for merging the regions, so that a final segmentation result is obtained. In the process of calculating the similarities among the regions, the color information is used, besides, the depth information is dynamically fused, and the problem that when the target color and the background color are similar, and namely a long-scale contrast edge exists between objects, correct segmentation can not be achieved is solved. The seed regions are automatically selected by the utilization of the depth information of the image, the seed regions of the target and the seed regions of the background do not need to be marked manually and alternately, region characteristics of the depth image are directly used for determining the seed regions instead of edge characteristics, and therefore high robustness is achieved.

The invention discloses a recognition and counting method for cells, comprising nine steps: preprocessing an image in a micron-order microscopic acquisition environment; extracting cell holes from the preprocessed image; performing closed hole filling of cells by using the knowledge of a connected domain; extracting a contour point sequence of cells from the image filled in step 3; filling non-closed holes of cells by adopting a non-closed hole filling method based on circularity determination; performing chamfer distance transformation on the filled image; performing extreme value uniqueness marking on cell hole positions; segmenting the image after extreme value uniqueness by using a marked watershed method; and quantifying and marking the segmented result. The method has the advantages that the influence of image noise can be greatly reduced, the phenomena of over-segmentation and discontinuous segment lines are eliminated, the segmentation effect is improved, and the cell recognition rate is improved.

The invention discloses an automatic segmentation method of a knee joint cartilage image. The method is characterized by comprising edge positioning based on SVM (Space Vector Modulation) and image segmentation based on a region growing method, wherein the step of edge positioning based on the SVM comprises acquisition and conversion of a knee joint MRI (Magnetic Resonance Imaging) image, adaptive Canny edge detection and cartilage edge classification based on the SVM; and in the step of image segmentation based on the region growing method, cartilage tissues are segmented by mainly adopting the improved region growing method capable of automatically selecting seed points. The method has the beneficial effects that the cartilage segmentation is performed on the knee joint MRI image; the precision positioning is realized by effectively combining the mode recognition with the edge detection; and the positioning complementation is sufficiently implemented in combination of the region growing method, so that the internal similar characteristics and the external difference characteristics of regions to be segmented are combined. Thus, the defects of result over-segmentation or inaccurate segmentation and the like of traditional segmentation method are effectively overcome.

The invention provides a method for segmenting adhesion cells in an image, which comprises the following steps of: performing image binaryzation, and performing distance transform on a binary image to obtain a distance value of each point; detecting local extreme points in a distance image, and screening to obtain central points representing cells; performing water area segmentation on the distance image, and determining the subjection relation of subblocks and the cells according to the distance between the center of each subblock and the center of each cell; and finally segmenting intermediate blocks adjacent to a plurality of subblocks accurately. By the method, the problems of over-segmentation and segmentation error which are likely to occur in the conventional watershed segmentation are solved effectively, and the method has the good applicability to the adhesion of the cells.

The invention discloses a multi-band high-resolution remote sensing image segmentation method based on a gray scale co-occurrence matrix. In texture images based on the gray scale co-occurrence matrix, precipitation watershed transformation is adopted for segmentation of each band image, and therefore band segmentation results are superposed. At last, a region merging strategy based on multi-band spectral information is provided for merging fragment regions in the segmentation results, and finally image segmentation is achieve. A high-resolution ALOS image and an SPOT 5 image are tested respectively through the method which is compared with a traditional segmentation method based on local area homogeneity gradients. As is shown in experimental results, edges of an object can be accurately positioned, over-segmentation and under-segmentation are effectively overcome, and higher segmentation accuracy and stability are achieved.

The invention provides a remote sensing image segmentation and identification method based on superpixel marking. Superpixel segmentation results are obtained by performing over-segmentation on remote sensing images by use of a superpixel segmentation algorithm, and learning samples are obtained by performing classification marking on superpixel blocks. Then, visual features of superpixel samples are extracted, these learning samples are trained by taking marking results as teacher signal classifiers, and trained classifier information is stored. The superpixel results are obtained by performing the over-segmentation on the remote sensing images to be analyzed, a visual feature of each superpixel is extracted and then is sent to the classifiers for classification, and after each superpixel block obtains a class mark, the superpixel blocks of the same class marks are merged, i.e., all areas of the images to be analyzed obtain class information. According to the invention, the remote sensing images are prevented from being directly segmented, edge information of actual areas is reserved to a quite large degree, segmentation and identification processes are integrated together, and the segmentation and identification capabilities are more excellent.

The invention discloses a super-pixel polarimetric SAR land feature classification method based on sparse representation. The method comprises: inputting polarimetric SAR image data to be classified, processing the image, and thereby obtaining a pseudocolor image corresponding to Pauli decomposition; performing super-pixel image over-segmentation on the pseudocolor image to obtain a plurality of super-pixels; extracting features, which are seven-dimensional, of radiation mechanism of the original polarimetric SAR image as features of every pixel; performing super-pixel united sparse representation to obtain sparse representation of each super-pixel feature; classifying by using a sparse representation classifier; working out the mean value of each super-pixel covariance matrix, then performing super-pixel complex Wishart iteration by using the classifying result in the last step, and at last obtaining a final classifying result. According to the super-pixel polarimetric SAR land feature classification method based on sparse representation, the problem that traditional classifying areas based on the single pixel are poor in consistency is solved, and operating speed of the algorithm is greatly increased on basis of improving accelerate.

The invention discloses a high-resolution remote sensing image segmentation method based on inter-scale mapping. The high-resolution remote sensing image segmentation method has the advantages that aiming at geographical object extraction involved in objet-level change detection, the high-resolution remote sensing image multi-scale segmentation method based on wavelet transform and an improved JSEG (joint systems engineering group) algorithm is provided; aiming at a key problem that a conventional JSEG algorithm affects segmentation accuracy in high-resolution remote sensing image segmentation, a corresponding improvement strategy is adopted to achieve a good effect; wavelet transform is introduced to serve as a multi-scale analysis tool, excessive rough color quantization of the conventional JSEG algorithm is abandoned, and thus, detail information in original images is kept to the utmost; a novel inter-scale segmentation result mapping mechanism is set up, image segmentation of a current scale is realized on the basis of a segmentation result of a previous scale, and correction of the segmentation result of the previous scale is also realized, and thus, accumulation of inter-scale segmentation errors is decreased effectively; finally, an improved multi-scale segmentation strategy and an improved region merging strategy are provided, and over-segmentation and mistaken merging are effectively reduced.

The invention discloses an image scene labeling method based on a conditional random field and a secondary dictionary study, comprising steps of performing superpixel area over-segmentation on a training set image, obtaining a superpixel over-segmentation area of each image, extracting the characteristics of each superpixel over-segmentation area, combining with a standard labeled image to construct a superpixel label pool, using the superpixel label tool to train a support vector machine classifier to calculate superpixel unary potential energy, calculating paired item potential energy of adjacent superpixels, in virtue of global classification statistic of the over-segmentation superpixel area in a training set, constructing a classifier applicable to a class statistic histogram as a classification cost, using the histogram statistic based on the sum of the sparse coders of the sparse representation of the key point characteristic in each class superpixel area as the high order potential energy of a CRF model, using two distinguishing dictionaries of a class dictionary and a shared dictionary to optimize the sparse coder through the secondary sparse representation, and updating the dictionary, the CRF parameters and the classifier parameters. The image scene labeling method improves the labeling accuracy.

The invention discloses a target tracking method based on dynamic graph matching, which is characterized in that a structural model of a dynamic graph is used for effectively tracking a target. The target tracking method mainly comprises six steps of superpixel over-segmentation, extraction of candidate target blocks, creation of an undirected graph, matching of the undirected graph, determination of a target position and updating of a model. The target tracking method has the advantages that the internal structural information of the target can be effectively modeled by using a target tracking model structured with the method, so that the target position can be still precisely determined by the tracking method when the target has greater structural deformation or is severely shielded.

The invention discloses a medical image Graph Cut segmentation method based on a statistical shape model. The medical image Graph Cut segmentation method mainly solves the problem that low-contrast organs cannot be effectively segmented in a medical image in the prior art, and is implemented by the steps of: (1) establishing the statistical shape model of the low-contrast organs, and acquiring grey scale information; (2) pre-segmenting the low-contrast organs; (3) initialization Graph; (4) and segmenting the low-contrast organs. The medical image Graph Cut segmentation method based on the statistical shape model adds priori knowledge of shapes of organs on the basis of rapid segmentation by adopting a Graph Cut algorithm, reduces the possibilities of over-segmentation and under-segmentation, determines initial positions of the low-contrast organs according to relative relationship between animal organs and animal in-vitro contours, improves the segmentation efficiency, and is a rapid and effective organ segmentation method.

The invention provides a multi-scale multi-level image segmentation method based on a minimum spanning tree. Multi-scale multi-level image segmentation is expressed and realized by a graph model. The segmentation method is suitable for initial segmentation results obtained by various images and various rules to effectively combine over-segmented regions in high level segmentation, so an over-segmentation phenomenon is avoided; moreover, multi-scale segmentation results on different levels provides characteristic description information on different levels for analysis of target structural components; and the method is very important for target recognition.

The invention discloses an automatic extraction method for the urban road network information of a high resolution remote sensing image. The automatic extraction method is characterized in comprisingthe following steps that: S1: on the basis of an improved watershed segmentation algorithm, selecting a proper local homogeneity threshold value, rejecting a local minimum small-area region, removingsmall plaques, and carrying out region combination to solve an over-segmentation problem; S2: in an object-oriented method, adopting geometrical characteristics and context characteristics, utilizingurban road network characteristics to extract a road image object and process occlusion problems in an image; S3: quickly and accurately extracting the position of a road intersection through an automatic extraction method of the road intersection in the high resolution image, and providing a basis for the topological connection of a road; and S4: aiming at the problems, including holes in surfaceshaped roads, fracture among road sections and the noise of "same spectrum with different objects" to lay a foundation for the topological connection of road extraction, and adopting two topologicalconnection methods to effectively connect road interruptions among road strips and road intersection positions for the large road interception in the road extraction result to further perfect the roadnetwork information.

The invention discloses an image annotation method combined with an image internal space relation and a visual symbiosis relation. The method consists of image segmentation, characteristic extraction and an annotation algorithm, and comprises the following steps of: firstly, segmenting an image into a plurality of regions by using an over-segmentation method; secondly, extracting the visual characteristics of each region; and lastly, establishing an image annotation classifying model by using context information such as a space position relation, the visual symbiosis relation and the like among regions in an image. The image annotation method has the advantages that: the image annotation accuracy is high, and the accuracy of image annotation can be increased by fully and effectively usingtwo kinds of different context information of the space position relation and the visual symbiosis relation in the image.

The invention provides a multiscale image natural color fusion method based on over-segmentation and optimization, which comprises the steps of: acquiring a grayscale fusion image; selecting a reference image, and carrying out over-segmentation treatment on the grayscale fusion image and the reference image; transforming the images from a RGB (Red Green Blue) color space to a 1alpha beta color space for area transmission; transmitting color information in an over-segmentation area in the reference image to the grayscale fusion image; determining a special area according to a point with the brightness higher than the integral image brightness in an infrared image; and transforming from the 1alpha beta color space to the RGB color space so as to acquire a fusion image with natural colors. The invention provides a multiscale image natural color fusion device based on over-segmentation and optimization, which comprises a transformation module, an over-segmentation module, an area transmission module, a color transmission module and a color fusion module. The method and device of the invention can improve the fusion effect of a multi-source grayscale image, enhance information resolving power, alleviate asthenopia of an observer and enhance sustainable observing ability.

The invention discloses an image scene segmentation and layering joint solution method based on component set sampling. The method comprises the following steps of: performing over-segmentation treatment on an input image to obtain a super-pixel set of the image; training on a training dataset to obtain a discriminant model of semantic category and a discriminant model of layer category, and obtain a probability value (of each super-pixel in the input image) belonging to each semantic category and a probability vale belonging to each layer category according to the two models; structuring a candidate graph structure of the input image and calculating a node weighted value, a positive side weighted value and a negative side weighted value; and based on the a candidate graph structure, obtaining an optimal solution by reasoning via a component set sampling algorithm, wherein the optimal solution includes the exact semantic category and the exact layer category of each super-pixel of the input image. The method disclosed by the invention can be widely used for semantic information and layer information labeling of computer visual systems of military, aviation, aerospace, monitoring and manufacturing, and the like.