41 results about "Unsupervised segmentation" patented technology

A method comprising: obtaining a three-dimensional (3D) point cloud about at least one object of interest; detecting ground and / or building objects from 3D point cloud data using an unsupervised segmentation method; removing the ground and / or building objects from the 3D point cloud data; and detecting one or more vertical objects from the remaining 3D point cloud data using a supervised segmentation method.

An image segmentation method includes generating a hierarchy of regions by unsupervised segmentation of an input image. Each region is described with a respective region feature vector representative of the region. Hierarchical structures are identified, each including a parent region and its respective child regions in the hierarchy. Each hierarchical structure is described with a respective hierarchical feature vector that is based on the region feature vectors of the respective parent and child regions. The hierarchical structures are classified according to a set of predefined classes with a hierarchical classifier component that is trained with hierarchical feature vectors of hierarchical structures of training images. The training images have semantic regions labeled according to the set of predefined classes. The input image is segmented into a plurality of semantic regions based on the classification of the hierarchical structures and optionally also on classification of the individual regions.

An image segmentation method based on a hierarchical higher order conditional random field model is provided, comprising: firstly, extracting a multi-class texture feature for a target image, and constructing a one-variable potential function and a pairing potential function of a pixel level; then, acquiring super pixel fragments of different granularities by using a unsupervised segmentation algorithm; designing a one-variable potential function and a pairing potential function of a super pixel level corresponding to each granularity layer; constructing a hierarchical higher order conditional random field model; learning a parameter of the hierarchical higher order conditional random field model in a supervision manner by using a manual marked sample; and finally, acquiring a final segmentation marking result for a to-be-tested image by means of model reasoning. The hierarchical higher order conditional random field model used in the present invention fuses multi-feature texture information and multi-layer super pixel segmentation information of an image, and can effectively improve boundary segmentation accuracy of a multi-target object in the image.

Provided herein are systems and computer-implemented methods for quantitative analyses of tissue sections (including, histopathology samples, such as immunohistochemically labeled or H&E stained tissue sections), involving automatic unsupervised segmentation of image(s) of the tissue section(s), measurement of multiple features for individual nuclei within the image(s), clustering of nuclei based on extracted features, and / or analysis of the spatial arrangement and organization of features in the image based on spatial statistics. Also provided are computer-readable media containing instructions to perform operations to carry out such methods. A quantitative image analysis pipeline for tumor purity estimation is also described

An image segmentation method includes generating a hierarchy of regions by unsupervised segmentation of an input image. Each region is described with a respective region feature vector representative of the region. Hierarchical structures are identified, each including a parent region and its respective child regions in the hierarchy. Each hierarchical structure is described with a respective hierarchical feature vector that is based on the region feature vectors of the respective parent and child regions. The hierarchical structures are classified according to a set of predefined classes with a hierarchical classifier component that is trained with hierarchical feature vectors of hierarchical structures of training images. The training images have semantic regions labeled according to the set of predefined classes. The input image is segmented into a plurality of semantic regions based on the classification of the hierarchical structures and optionally also on classification of the individual regions.

The invention relates to a fabric defect detection and classification method based on unsupervised segmentation and ELM (Extreme Learning Machines), and belongs to the fields of computer vision, pattern recognition and image application. Fabric defect images are input, and fabric defect image segmentation is realized to obtain segmented fabric defect images; shape and texture features of the segmented defect images and the original defect images are extracted to realize feature extraction of the fabric defect images; a feature set and labels of the to-be-classified fabric defect images are used as a training set to train an ELM classifier to obtain ELM classifier parameters; and fabric defect image classification is realized through the trained classifier according to Bayesian probability fusion, and a fabric defect image classification result is output. The detection and classification method provided by the invention has a higher accuracy rate.

The invention discloses an SAR image segmentation method based on semantic information classification. The SAR image segmentation method based on the semantic information classification mainly solves the problem that ground object zones, formed by uniformly connective ground object target gathering, of a forest, a building group and the like can not be obtained through non-supervision segmentation by an existing segmentation method. The method comprises the following steps: (1) an initial sketch model is used on an input SAR image so that an initial sketch image expressing image structure information is obtained; (2) semantic information analysis is performed on the initial sketch image so that semantic information classification results of all line segments are obtained; (3) the ground object zones formed by the ground object target gathering are classified based on the semantic information analysis; and (4) the rest zones are divided into zones to be determined and non-line-segment zones and SAR image segmentation is respectively performed to the zones to be determined and the non-line-segment zones so that the SAR image segmentation is finally achieved. Compared with the prior art, the SAR image segmentation method based on the semantic information classification is strong in generality and capable of achieving segmentation of SAR images with a large amount of ground object zones formed by the ground object target gathering. Uniform connectivity of a segmentation result is good, edge location is accurate, and the independent ground object target can be segmented.

The invention discloses a stomach computer tomography (CT) image lymph node automatic auxiliary detecting system without supervision segmentation. The stomach CT lymph node automatic auxiliary detecting system without supervision segmentation mainly aims at solving the problems that an area-of-interest and a suspected lymph node area in an existing stomach CT image automatically acquire undetected areas or excessive irrelevant information is left over. The stomach CT lymph node automatic auxiliary detecting system without supervision segmentation comprises a preprocessing module, an area-of-interest extracting module, a suspected lymph node extracting module and a lymph mode tracking extracting module. The preprocessing module is used for preprocessing to-be-detected images, the area-of-interest extracting module is used for further processing the preprocessed images to acquire areas-of-interest, the suspected lymph node extracting module is used for extracting suspected lymph nodes out from the areas-of-interest, and the lymph node tracking extracting module is used for window feature matching tracking of the suspected lymph nodes, and completing mark extraction of the lymph nodes. The stomach CT lymph node automatic auxiliary detecting system without supervision segmentation is capable of automatically and effectively extracting suspected lymph node areas which draw attentions of doctors and eventually detecting the lymph nodes, and being used for processing medical images.

The invention provides a mean shift and fuzzy clustering-based natural image unsupervised segmentation method, and mainly solves the problem of low unsupervised segmentation accuracy of massive natural images in the prior art. According to the scheme, the method comprises the steps of 1) inputting an image and smoothing the image; 2) uniformly initializing 64 iterative initial points in a normalized RGB color space of pixels of the smoothed image; 3) performing an iterative search on the initial points to obtain 64 convergence points; 4) deleting the convergence points with pixel numbers smaller than a deletion threshold in high-dimensional balls taking the convergence points as centers; 5) combining the convergence points with Euclidean distances smaller than a combination threshold, determining density peak values and a density peak value number, and calculating membership degrees of the pixels and smooth membership degrees of the pixels in sequence; and 6) defuzzifying the smooth membership degrees of the pixels, adding class tags to the pixels, and outputting segmented images. According to the method, control parameters do not need to be set; a segmentation type number of the image can be automatically determined; and the method can be used for unsupervised segmentation of the massive natural images.

The invention relates to a computing device-implemented method and apparatus for unsupervised segmentation of microscopic color image of unstained specimen and digital staining of segmented histological structures. Image of unstained specimen is created by light microscope 101, recorded by color camera 102 and stored on computer-readable medium 103. The invention is carried out by a computing device 104 comprised of: computer-readable medium for storing and computer for executing instructions of the algorithm for unsupervised segmentation of microscopic color image of unstained specimen and digital staining of segmented histological structures. Segmented histological structures and digitally stained image are stored and displayed on the output storing and display device 105 in order to establish diagnosis of a disease. The invention is an improvement over the prior art as it is characterized by the: (i) shortening of slide preparation process; (ii) reduction of intra-histologist variation in diagnosis; (iii) elimination of adding chemical effects on specimen; (iv) elimination of altering morphology of the specimen; (v) simplification of histological and intra-surgical tissue analysis; (vi) being significantly cheaper than existing staining techniques; (vii) being harmless to the user because toxic chemical stains are not used; (viii) discrimination of several types of histological structures present in the specimen; (ix) usage of the same specimen for more than one analysis.

The invention discloses an iteration-based three-step unsupervised Chinese word segmentation method and belongs to the field of natural language processing technology. According to the basic thought,the method is an unsupervised word segmentation framework including local segmentation, global word selection and corpus reduction iteration execution; and in each iteration, a word formation probability model based on segmentation-context mutual independency is utilized to perform locally optimal unsupervised segmentation on text corpus, and the form is simple and effective; a document-level pulse weighting method is adopted according to the long-tail phenomenon; according to a global support degree, new words are screened, and a dictionary is incrementally generated; and last, a text is divided based on the longest matching and maximum probability principle of the dictionary, formed segmented words are filtered out, continuous non-segmented words are stitched, the words are reconstructedinto a scale-reduced training corpus, and similar iteration processing is performed on the remaining corpus till no new word is generated. The method is superior to an existing Chinese unsupervised word segmentation algorithm with best performance.

The invention discloses a G0 distribution-based stochastic gradient variational Bayesian SAR image segmentation method. The method comprises the steps of extracting a sketch drawing of an SAR image according to an initial sketch model; according to a regional chart, dividing the SAR image into a mixed pixel sub-space, a uniform pixel sub-space and a structure pixel sub-space; for each extremely non-uniform region in the mixed pixel sub-space, estimating a G0 distribution parameter of each extremely non-uniform region, and learning structural features of each extremely non-uniform region by utilizing a G0 distribution-based stochastic gradient variational Bayesian model, thereby realizing unsupervised segmentation of the mixed pixel sub-space; and performing corresponding segmentation on the uniform pixel sub-space and the structure pixel sub-space, and fusing segmentation results of the three sub-spaces to obtain an SAR image segmentation result finally. According to the method, hidden variable prior distribution and similar posterior distribution in the model are both assumed to meet G0 distribution of extremely non-uniform regions, and a corresponding analysis formula is derived for performing learning, so that the accuracy of clustering the extremely non-uniform regions in the mixed pixel sub-space is improved.

A method comprising: obtaining a three-dimensional (3D) point cloud about at least one object of interest; detecting ground and / or building objects from 3D point cloud data using an unsupervised segmentation method; removing the ground and / or building objects from the 3D point cloud data; and detecting one or more vertical objects from the remaining 3D point cloud data using a supervised segmentation method.

The invention relates to an unsupervised medical image segmentation method based on an adversarial network, belongs to medical care informatics, and particularly relates to the technical field of medical image segmentation. According to the technical scheme, the method comprises the following steps: firstly, randomly generating or utilizing a third-party data set to obtain a group of auxiliary masks according to shape prior information, and sending the auxiliary masks and an unlabeled training image into a cyclic consistency adversarial network to generate binary masks; and a discriminator based on variational self-encoding and a generator correction module based on discriminator feedback are utilized to improve the quality of the binary mask. And after the binary mask of the training image is obtained, iterative training is performed by using a noise weighted Dice loss function, so that a final high-precision segmentation model can be obtained. According to the method, the problem that the convolutional neural network needs a large number of manual annotations in the training process of medical image segmentation can be solved, the problems of low performance, poor robustness and the like of an unsupervised segmentation method are solved, and the performance of an unsupervised medical image segmentation algorithm is effectively improved.

Provided is a behavior video non-supervision time-sequence partitioning method. The method comprises steps of: initializing the initial time of video detection as nt and a corresponding sliding window frame length as Lt; detecting behavior change points in an established video sequence window; if it is detected that a behavior change point c is in the video sequence window, using a time point c as the initial time of detection and reinitializing sliding window frame length in order to continue detecting subsequent videos; if it is not detected that the behavior change point is in the video sequence window, still using the nt as the initial frame of detection, namely nt+1=nt, and updating the sliding window frame length as Lt+1=Lt+[delta]L, wherein the [delta]L is a incremented step length of the length of the sliding window; ending the method until all video frame sequences are detected or reach a predetermined deadline T0. The method makes a decision on data change points in behavior video analysis, does not require online and real-time non-supervision partition with prior knowledge, and is directly used in behavior video data online analysis.

The invention discloses an image segmentation method based on a layered high-order conditional random field model, and the method comprises the steps: firstly extracting a plurality of types of texture features of a target image, and constructing a pixel-level unary potential function and a pixel-level paired potential function; obtaining superpixel segments with different granularities by using an unsupervised segmentation algorithm; designing a unary potential function and a paired potential function of each granularity layer corresponding to the superpixel level; constructing a layered high-order conditional random field model; supervising and learning hierarchical high-order conditional random field model parameters by utilizing the artificially marked samples; and finally, carrying out model reasoning on the to-be-tested image to obtain a final segmentation marking result. The layered high-order conditional random field model adopted by the invention fuses the multi-feature texture information and the multi-layer superpixel segmentation information of the image, so that the boundary segmentation accuracy of the multi-target object in the image can be effectively improved.

The invention discloses a domain adaptive unsupervised image segmentation method based on generative adversarial and class feature distribution. The method is characterized in that unsupervised semantic segmentation is carried out on a target domain image by adopting adversarial learning and a self-learning method based on class feature distribution, wherein the invention specifically comprises two networks of a main semantic segmentation network and an auxiliary semantic segmentation network, and two modules of a domain alignment adversarial module and a pseudo-label optimization module. Compared with the prior art, the method has the advantages of high segmentation precision and multi-granularity semantic consistency, unsupervised semantic segmentation of the target domain image is realized by utilizing adversarial learning and self-learning based on class feature distribution, the multi-granularity semantic consistency between the domains is enhanced by aligning the domains on global and class levels, and the problem that a traditional unsupervised segmentation method is insufficient in segmentation precision is solved.

The invention relates to the technical field of medical image analysis, and discloses an unsupervised segmentation method for multi-mode brain tumor MRI, which comprises the following steps of 1) inputting different modal images of a tumor patient, including T2 weighting, Flair and T1 enhanced images, and carrying out gray value normalization on the input images; and 2) extracting features of pixel points of the brain region on T2 and Flair images, classifying the pixel points by using a clustering fusion method, and automatically identifying the category of the tumor from a plurality of categories. According to the unsupervised segmentation method for the multi-mode brain tumor MRI, the imaging characteristics of different MRI modes are fully utilized; the information of different modes and neighborhood information are combined, so that the segmentation accuracy is improved; an unsupervised segmentation method is used, a large amount of labeled data is not needed, the long-time training time and complex calculation are not needed, so that a large amount of work is facilitated, the segmentation accuracy is effectively improved, and the purposes of being easy and efficient to implement and high in operation speed of the algorithm are achieved.

The invention discloses a lung disease lesion unsupervised segmentation method based on knowledge distillation, and belongs to the field of medical image processing and computer vision. The method comprises the following steps: firstly, constructing and training an auto-encoder to obtain a pre-trained teacher network with rich CT image semantic knowledge, then training a student network with the same architecture as the teacher network from the pre-trained teacher network by only distilling the knowledge of normal CT images, and finally, performing lesion segmentation on the difference of features extracted by the lesion-containing image by using teacher and student networks. Meanwhile, besides traditional pixel-level distillation, affinity-level distillation considering the relation between pixels is further designed in the method, so that effective knowledge is fully distilled. Experiments prove that the lesion segmentation precision can be effectively improved on different data sets. The method is easy to construct, an unlabeled lung disease focus segmentation result can be obtained only depending on normal data, and generalization and operation efficiency are considerable.

The invention relates to a moving workpiece target unsupervised segmentation method suitable for a high-dynamic light condition, which belongs to the technical field of digital image processing. Single frame superposition is carried out on acquired continuous frames of video images, and subtraction with a background image is carried out; then, an LDOF optical flow method is used for acquiring a video foreground moving target optical flow field, and according to the optical flow field, a foreground moving target boundary is acquired; and a mixed Gaussian model and a position model are built for the foreground and the background of the video image. Through the above processing, segmentation on the video foreground target is finally realized. In view of the defect that the segmentation precision of a moving target quick segmentation method in an unrestricted scene is likely influenced by a light condition, an image pre-processing method of adopting single frame superposition to make the feature information of the foreground moving target to be more outstanding is put forward, influences on the segmentation precision of the method by the light condition can be effectively reduced and the video segmentation precision is improved.

The invention provides an unsupervised domain adaptation method and system based on target domain self-supervised learning. The unsupervised domain adaptation method comprises a domain invariant feature extraction step, an image feature stepped domain alignment step, a semantic segmentation step, an edge generation step, a segmentation image domain alignment step, an edge image domain alignment step and an edge consistency constraint step. According to the method, effective self-supervised learning is carried out on the target field, so that the unsupervised segmentation precision of the target field is improved and good field adaptation is realized.

The invention belongs to the technical field of remote sensing image processing, and discloses a remote sensing image unsupervised segmentation evaluation method and device. The method comprises the steps of obtaining a plurality of segmentation objects of a remote sensing image; according to the area of each segmentation object, weighting the average variance of all wavebands of each segmentationobject to obtain an area weighted variance; obtaining the relative heterogeneity of each segmentation object according to the local spectrum difference and the average spectrum difference of each segmentation object; obtaining total heterogeneity according to the number of the segmentation objects, the waveband number of the remote sensing image and the relative heterogeneity of each segmentationobject; and obtaining an evaluation result according to the area weighting variance and the total heterogeneity. The device comprises an obtaining module, an area weighted variance obtaining module,a relative heterogeneity obtaining module, an overall heterogeneity obtaining module and an evaluation result obtaining module. According to the method, the over-segmentation or under-segmentation phenomenon caused by a traditional evaluation method can be avoided, so that the image evaluation method is optimized, the evaluation effect is more remarkable, and the evaluation precision is higher.

The invention discloses an unsupervised segmentation method for terracotta army point clouds based on combination of region growth and deep learning. The method comprises the steps of performing unsupervised semantic segmentation on point cloud data, performing normal vector prediction according to three-dimensional coordinate data of point cloud, and then performing pre-segmentation based on normal vectors on the point cloud; carrying out pre-segmentation on a point cloud by using SRG-Net, then carrying out partial segmentation on the point cloud by using SRG-Net, finally, automatically adjusting a segmentation result, and enhancing a result in the SRG-Net according to a pre-segmentation result in SRG so as to obtain a segmentation label of the point cloud. Unsupervised part segmentation of the terracotta warrior point clouds is realized, and a good effect can be obtained for segmentation of general point clouds except for the terracotta warrior point clouds.

The invention relates to a brain MRI image segmentation method combined with biological characteristics, belonging to the technical field of image processing and biomedical combination. The method ofthe present invention utilizes H-type biometric features of CSF central part to segment CSF part, the rest of the brain is segmented by the fuzzy C-means clustering algorithm based on local information as the initial label input of the random forest, the symmetry of the brain image is used, the left half pixels of the MRI image are used as the training samples of the random forest classifier, andall the pixels are used as the test samples of the random forest classifier. Finally, the final segmentation result of the brain MRI image is obtained by combining the segmentation results of the twoparts. The final result of the invention is better than other segmentation algorithms only considering image gray information, and the unsupervised segmentation without requiring a large number of picture training sets is realized.

The invention discloses a full convolutional neural network cultivated land extraction method and system based on multi-scale fusion, and the method comprises the steps of carrying out the unsupervised segmentation of a high-resolution remote sensing image based on an unsupervised multi-scale segmentation mode, carrying out the image processing of a plurality of obtained vector units, and obtaining a training sample data set; building a multi-scale fusion full convolutional neural network, using the training sample data set to train the multi-scale fusion full convolutional neural network, where the multi-scale fusion full convolutional neural network adopts four-scale feature network structures which are connected in parallel, and a feature output layer is added into a neighborhood feature fusion module; and inputting a cultivated land image to be extracted into the trained multi-scale fused full convolutional neural network for prediction, generating a category grid image, and obtaining a final cultivated land vector pattern spot through an image post-processing algorithm. According to the method, relatively regular and homogeneous cultivated land parcel vectors can be stably extracted from a high-resolution image.

The invention belongs to the technical field of image processing, and particularly relates to a method for replacing a color of an image based on color harmony. The method comprises the following steps: firstly, carrying out online crawler on an online website Colorpalettes to obtain all pictures as a data set; performing unsupervised segmentation on each image according to colors to obtain an initial segmentation image, an initial color block type of the initial segmentation image and the number corresponding to each color block; secondly, optimizing the initial segmentation map, the types of the initial color blocks and the number corresponding to each kind of color blocks; segmenting and optimizing the input image, matching a proper color block combination in a database, and giving three color replacement schemes; and finally, optimizing the color lump combination obtained by the three color replacement schemes according to the artistic theory of color matching, and completing color replacement on the original image. The method can adapt to images of any number of color themes in a scene.

The invention provides an automatic segmentation method for articular cartilage tissue in a three-dimensional medical image. The automatic segmentation method is composed of three parts of three-dimensional medical image preprocessing, articular cartilage tissue coarse segmentation and articular cartilage tissue correction. According to the automatic segmentation method for the articular cartilagetissue in the three-dimensional medical image, the three-dimensional medical image is input, and the corresponding articular cartilage tissue is output. The method is an unsupervised segmentation method, can quickly and automatically realize cartilage tissue segmentation in a three-dimensional medical image without any manually labeled training sample, and has segmentation precision equivalent tothat of a supervised segmentation method.

The invention provides a device for creating a three-dimensional model of an object in a luggage to be inspected which includes an acquisition module for acquiring tomographic data of the luggage to be inspected using a CT imaging system; a first processing module, configured to perform interpolation processing on the sectional data to generate three-dimensional volume data of the checked luggage;a second processing module, configured to perform a plurality of smoothing processes on the three-dimensional volume data; a third processing module, configured to perform unsupervised segmentation on the smoothed three-dimensional volume data to obtain a plurality of segmented regions; a fourth processing module is used for extracting the isosurface of the plurality of divided regions and obtaining the corresponding isosurface; a fifth processing module, configured to perform three-dimensional surface division on the isosurface to form a three-dimensional model of each object in the checkedluggage.

The invention relates to a blood leukocyte staining style conversion method based on a cyclic generative adversarial network. By introducing an unsupervised segmentation strategy, the problem that the cost is high when tag information of a leukocyte image is obtained as semantic condition limitation is solved, and existing image data are fully utilized; through an improved multi-scale discriminator, deep and shallow layer feature information obtained by multi-scale down-sampling operation is subjected to feature fusion correction through a self-correction module (SCBlock), so that the definition and semantic accuracy of a converted image are ensured. Wide experiments show that the converted leukocyte rapid staining image generated by the method is clear and accurate, and meanwhile, the leukocyte image segmentation performance can be greatly improved by utilizing the leukocyte rapid staining image generated by the method.

A method and apparatus for creating a three-dimensional model of an object are disclosed. The method comprises the steps of: using a CT imaging system to acquire tomographic data of checked luggage; interpolating the tomographic data to generate three-dimensional volume data of checked luggage; performing unsupervised segmentation on the three-dimensional volume data of checked luggage to obtain multiple divided area; perform isosurface extraction on multiple segmented areas to obtain the corresponding isosurface; perform three-dimensional surface segmentation on the isosurface to form a three-dimensional model of each object in the checked luggage. Using the above solution, the three-dimensional model of the object in the checked baggage can be established more accurately, thereby providing a better basis for subsequent shape feature extraction and security inspection, thereby reducing missed inspections.