67results about How to "Solve the segmentation problem" patented technology

The embodiment of the invention discloses an image feature extraction method and a device thereof, wherein, the image feature extraction method comprises the steps as follows: a weighted image is generated according to a collected image and a pixel of the weighted image is obtained from weighting the sum-difference ratio of a corresponding pixel of the pixel of the weighted image in a RGB color space and the gray value of a corresponding pixel in a Lab color space; a binary image is generated according to the weighted image and comprises a target image and a background image; a hole of the target image in the binary image is filled and noise in the binary image is also removed; edge images of the target image after filling and contour-rectangle images of the edge images are both obtained; and as to any contour-rectangle image, the center and radius of a corresponding circle is obtained according to the calculation of the edge image in the contour-rectangle image. The image feature extraction method and the device thereof proposed solve problems in the recognition of a target-segmented image when targets are more, overlapped with each other and blocked, and are characterized by accurateness and high speed.

The invention discloses an unsupervised cervical cell image automatic segmentation method. The method comprises the following steps of 1) preprocessing a cervical cell image; 2) subjecting the preprocessed cell image to the pre-background rough segmentation, and extracting a region to which cells belong; 3) subjecting the roughly segmented cell image to the detection and segmentation of cell components, and segmenting cells of different types by using a rapid region convolution neural network; 4) detecting and segmenting the cell nucleuses of cervical cells; 5) according to the characteristic parameters of the cell nucleuses, screening the cell nucleuses to obtain final candidate cell nucleuses; 6) judging whether the cell types obtained in the step 3) are multicellular spheroids or not; if not, segmenting a cytoplasmic region by using an activity contour model and a prior template; otherwise, conducting the step 7); 7) conducting the post-treatment based on the segmentation result of cell nucleus and cytoplasm, and the domain knowledge. In this way, the effective segmentation of a whole cervical cell is completed.

The invention discloses a blood vessel segmentation method and device. The method includes: using sample image training to obtain a blood vessel central line average model; aligning a to-be-segmented image with the template image of a sample image; mapping the blood vessel central line average model to the to-be-segmented image; regulating the blood vessel central line average model on the basis of the priori knowledge of the to-be-segmented image to obtain the blood vessel central line of the to-be-segmented image; performing blood vessel segmentation on the to-be-segmented image on the basis of the blood vessel central line of the to-be-segmented image. By the method, blood vessel segmentation problems can be well solved, blood vessel segmentation is clean, and segmentation failure caused by the fact that the bone is segmented.

An active contour model based method for segmenting mammary gland DCE-MRI focus belongs to the field of medical image segmentation and comprises the following steps: obtaining mammary gland DCE-MRI image sequence data by MRI scanning equipment; manually selecting a region of interest; automatically obtaining subtracted size of interest, active contour segmenting focus and visually display focus. According to the invention, based on the features that statistical distributions of mammary gland DCE-MRI image backgrounds are consistent and internal distributions in the focus are different, an edge stopping function of the active contour model is designed, thereby realizing reliable segmentation of the focus and effectively avoiding edge outleakage phenomenon; during the model evolutionary process, re-initialization of a signed distance function is not required, so that the real-time performance of the system is higher. The method has a lower requirement on manual operation in implementation, is high in intelligent degree, low in data storage space requirement, and quick in processing speed, and can effectively obtain comprehensive and steric space information of the focus through three-dimensional angle segmentation, which facilitates the multi-angle observation and analysis of the focus by a doctor.

The invention relates to a classifying method of three-dimensional point-cloud data, in particular to a feature ball based classifying method of three-dimensional point-cloud data of an outdoor scene. The method includes the steps of 1, constructing a conditional random field model; 2, constructing a three-dimensional point-cloud feature ball; 3, calculating point feature vectors; 4, calculating edge feature vectors; 5, calculating group feature vectors; 6, learning parameters of the conditional random field model; and 7, subjecting the three-dimensional point-cloud data to inference classification. The method has the advantages that feature vectors of all levels of three-dimensional point clouds are accurately and comprehensively calculated through construction of the three-dimensional point-cloud feature ball, the three-dimensional point cloud of the outdoor scene is accurately and reliably segmented, point cloud groups uniform in property are formed, the problem that imperfect structure of the point cloud feature vectors and inaccuracy of point cloud segmentation caused by the factors such as complex geometric topological structure of the outdoor scene are effectively solved, and the effect of classified recognition of the three-dimensional point-cloud data of the outdoor scene is greatly improved.

The invention relates to a texture surface defect detection and segmentation device and method in an industrial environment, and belongs to the technical field of industrial texture surface defect detection and segmentation. The device comprises an input module I, a feature extraction module II, a defect region extraction module III, a defect region alignment module IV, a detection module V and asegmentation module VI; the input module I outputs to the feature extraction module II; the feature extraction module II outputs to the defect region extraction module III and the defect region alignment module IV, the defect region extraction module III outputs to the defect region alignment module IV, and the defect region alignment module IV outputs to the detection module V and the segmentation module VI. The specific types, accurate position coordinates and spatial distribution information of multiple types of defects can be automatically obtained, and the method has good robustness for the multi-scale of the defects. The method has wide applicability and relatively high detection precision for texture surface defects.

The invention relates to a road scene segmentation method based on a full convolutional neural network, and the method comprises the following steps: 1, carrying out the median filtering of an original road scene image through a KSW two-dimensional threshold value and a genetic algorithm, and obtaining a training set; 2, constructing a full convolutional neural network framework; 3, taking a training sample obtained at step 1 and an artificial segmentation image discriminated and identified through human eyes as the input data of the full convolutional neural network, and obtaining a deep learning neural network segmentation model with the higher robustness and better accuracy through training; 4, introducing to-be-segmented road scene image test data into the trained deep learning neuralnetwork segmentation model, and obtaining a final segmentation result. An experiment result indicates that the method can effectively solve a segmentation problem of a road scene image, has the higherrobustness and segmentation precision than a conventional road scene image segmentation method, and can be further used for the road image segmentation in more complex scenes.

The present invention relates to an object-learning robot and corresponding method. The robot comprises a gripper (14) for holding an object (11) to be learned to the robot (10); an optical system (16) having a field of view for introducing the object (11) to the robot (10) and for observing the gripper (14) and the object (11) held by the gripper (14); an input device (26) for providing an object identity of the object to be learned to the robot (10); a controller (24) for controlling the motion of the gripper (14) according to a predetermined movement pattern; and an image processing means (28) for analyzing image data obtained from the optical system (16) identifying the object (11) for association with the object identity. This enables the robot to learn the identity of new objects in a dynamic environment, even without an offline period for learning.

The invention discloses a similarity data clustering method for dam safety monitoring data. The method comprises the following steps of separating a single measuring point sequence trend term from thehigh-frequency noise by utilizing an EMD algorithm, detecting the time sequence change points by adopting an inflection point detection method of a cumulative sum control graph, and splitting to obtain all subsequence sets; adopting a DTW distance measurement method for calculating the distance problem of the subsequence, and calculating the distance minimum value between the two pieces of subsequence data dynamically; and clustering the mined sub-time sequences by using hierarchical clustering, and dynamically analyzing the time sequence clustering distribution condition under different clustering numbers through the obtained tree-shaped clustering graph. According to the method, the similarity of the monitoring data is reasonably analyzed, the correlation of the same monitoring point inthe time sequence can be mined, and meanwhile the correlation between the safety monitoring data can be quantified. And the monitoring data subjected to similarity analysis processing can accuratelyreflect the change trend of the dam in the time dimension, and the subsequent monitoring data mining difficulty can be effectively reduced in combination with the change trend rule.

The invention discloses a background frame feature-based adhesion rice grain image segmentation method and belongs to the image segmentation field. The method includes the following steps that: the binary image of a rice grain adhesion region is obtained by means of image preprocessing and image connected region shape analysis; the minimum convex hull image of the rice grain adhesion region is calculated, and a background region image within a minimum convex hull range is extracted, and frame extraction processing is performed on the obtained background region image, and frame end points are obtained on the basis of a frame feature image, and are adopted as segmentation end points; and optimal segmentation positions are obtained by means of by marching and retrieval on the basis of end point matching rules, and segmentation curves are drawn, so that an adhesion rice grain segmentation result can be obtained. The method of the invention can obtain higher accuracy and boundary smoothnessof adhesion rice grain image segmentation. When being applied to a rice grain quality automatic detection system, the method can effectively improve detection precision.

The invention belongs to the technical field of medical early warning, and discloses an early warning system and method for cerebral apoplexy. The early warning system for cerebral apoplexy comprisesan image collection module, a physiological index collection module, a main control module, an image processing module, a risk evaluation module, a warning module, a data storage module and a displaymodule. According to the system and method, through the image processing module, automatic identification and segmentation of a cerebral ischemia zone in a super-acute stage can be achieved, and the segmentation precision of the ischemia zone is improved; data preprocessing, feature selection and feature optimization are conducted through the risk evaluation module, and obtained data features aremore effective; an XGBoost method is adopted for automatically generating the risk probability that a target group of people suffer from cerebral apoplexy, community health screening can be more efficiently and conveniently conducted, and a doctor can be helped to more simply and quickly evaluate the risk that the target group of people suffer from cerebral apoplexy; health screening can be strongly pushed, so that potential users with cerebral apoplexy are more quickly found, and earlier warning is given, so that earlier and more effective treatment is conducted.

The invention belongs to the technical field of laser etching, and discloses a pattern segmentation method for laser etching processing, and a laser etching control system, and the method comprises the steps: calculating a normal vector of a mapping point; winding the two-dimensional vector maps on the surface of the three-dimensional grid model, calculating a transformation matrix of each two-dimensional vector map, and winding the maps on the surface of the three-dimensional grid model based on a geometric transformation principle; determining a search range according to the scanning range of the scanning galvanometer and the maximum focal depth, and obtaining a mapping pattern associated with each processing area; for one processing area, determining a segmentation plane according to the position of the central point, projecting the boundary line of the processing area and the associated pattern onto the segmentation plane, and segmenting the pattern; and according to the back projection coordinate system, generating a machinable file required by the machining party. The moving times of the scanning galvanometer in a traditional method can be effectively reduced, and the actualmachining efficiency is improved; and the pattern of each processing area can be efficiently segmented, and meanwhile, the precision of the pattern is ensured.

The invention discloses a bar reviewing-counting method and device. Bundled bars can be automatically reviewed and counted by the method and device, and counting can be more precise and faster. The method has the advantages that the technical problem that background interference occurs in the site of end partitioning, feature recognition and removal detection for the bundled bars is solved in terms of image processing, and counting precision is improved; a high-brightness LED light source is used in the reviewing-counting device is used to enhance detection effect, an industrial camera is used to acquire images for the ends of bars, an industrial personal computer is used for smooth-filtering and setting a threshold for binary treatment, corrosion and connected domain treatment, counting and comparative judgment are achieved, counting results are output by a field display, and the device has the advantages of high speed, high adaptability, simplicity and reliability and the like; production efficiency is improved, labor intensity is reduced for workers, and the method and device has great practical value.

The invention relates to the technical field of multimedia information processing, and provides a video reorganization system capable of automatically partitioning a shot and a video reorganization method thereof. The video reorganization system is as follows: a partition processor is respectively connected to an original video library and a reorganizer; and a generation and storage module is connected with the reorganizer. The video reorganization method comprises the following steps that: a user opens the original video library and extracts a video in need of being processed from the original video library; the partition processor performs automatic shot partition of the video, such that an independent video processing unit is obtained; a single-video effect library, a replacement background module and a multi-video operation module of the reorganizer are used for correspondingly processing a video processing unit respectively; and the processed video is transmitted to the generation and storage module and then stored and played. The video reorganization system disclosed by the invention is convenient to install and low in system hardware requirement and is capable of realizing operations, such as automatic shot partition, single-video effect processing, background replacement and multi-video mixing, on the same platform; therefore, the video processing efficiency of the video reorganization system is ensured; and thus, the video reorganization system has perfect functionality.

The invention discloses a pavement segmentation method and device based on a multi-line laser radar, and the method comprises the steps: S1, scanning the surrounding environment of a vehicle through the multi-line laser radar, obtaining laser radar point cloud data under a laser radar coordinate system, and converting the laser radar point cloud data into laser radar point cloud data under a vehicle coordinate system; S2, projecting the laser radar point cloud data under the vehicle coordinate system into a two-dimensional grid map; S3, according to the two-dimensional grid map, carrying out road surface segmentation based on a local accumulated height difference and a regional laser beam distribution threshold value; and S4, superposing road surface segmentation results based on the local cumulative height difference and the regional laser beam distribution threshold to obtain a complete road surface segmentation result. According to the method, the local feature fusion method is adopted, the road surface segmentation problem of the automatic driving automobile under the complex environment working condition is effectively solved, the steps are simple and easy to implement, and the method has high robustness and real-time performance.

The invention discloses a binary tree-based text line accurate locating method. The method comprises the steps of 1, constructing a binary tree search space for an over-segmented text line for an input natural scene image with the over-segmented text line; 2, simulating an access mode of postorder traversal in the constructed binary tree search space to search for paths, wherein each path is a node set, and the optimal path is a node set sequence obtained after all node sets are subjected to pruning and fusion operations; and 3, performing the pruning and fusion operations: for each node set contained in a path of the binary tree search space, a proper situation can be selected from eight pruning and combination policies to perform processing according to a confidence degree comparison result of a father node and left and right sub-nodes in the node set; and finally obtaining the optimal path which solves the over-segmentation problem, wherein the optimal path corresponds to a reset text line accurate locating result. According to the method, accurate locating of the text line in the scene image is realized.

The invention relates to the field of image segmentation, and particularly relates to an active contour image segmentation method and device based on improved SPF. According to the method, firstly, asign pressure function is redefined based on the global information of the image, and then a local image fitting LIF model is introduced, finally, a weight function model is established based on imagelocal and global information, the self-adaptive adjustment of the weights between the local information items and the global information items is achieved, and a new level set evolutionary equation is obtained. According to the method, the problem that the gray non-uniform image cannot be accurately segmented by only utilizing the image global information is solved, and the defect that a model based on the image local information is sensitive to noise and initial contour is overcome. The method is simple in calculation and high in convergence speed, and can effectively solve the problem of segmentation of multi-target and gray non-uniform images, and meanwhile, the robustness for the initial contour and the noise is relatively high.

The invention discloses an image segmentation method combining thermodynamic diagram channel and improved U-Net. The method comprises the following steps: 1) obtaining an original image; 2) preprocessing the original image to obtain a thermodynamic diagram; 3) constructing a multi-scale convolution module to improve a U-Net network structure; 4) entering a thermal map prepared in step 2) into theimproved U-Net network structure for image segmentation to obtain the segmented image. The multi-scale convolution module and a thermodynamic diagram channel are added to the U-Net network structure.Compared with a traditional U-Net model, the multi-scale convolution module and the thermodynamic diagram channel are added, so that not only can complete feature information be obtained, but also theprecision of a target edge can be effectively improved, and the edge is smoother.

The invention discloses a sign language image segmentation method based on centroid positioning and distance transformation. The method comprises the following steps: firstly, converting a read sign language image into a YCbCr color space, and performing skin color clustering; Performing smoothing processing on the segmented sign language image by adopting median filtering, performing maximum threshold binaryzation on the filtered sign language image, and finally performing processing by adopting an image region filling algorithm to ensure the integrity of a sign language region in the image;performing Processing by adopting area operator and centroid positioning to obtain a hand-arm region; separating the hand-arm region by the distance-based image based on the distance transform method,so as to achieve accurate segmentation of the hand region and the arm region. The method provided by the invention solves the problem of segmentation of the arm area, is suitable for segmentation oftwo conditions of a large bent arm and a long arm, has good robustness, can adapt to the two arm conditions, and can also adapt to gesture rotation and gesture size change.

The invention belongs to the field of electric logging value simulation, and discloses a method for simulating fracture-vug type reservoir body electric logging response. The method includes establishing an equivalent fracture-vug type reservoir body stratum model according to regional geological data, determining basic parameters of the equivalent fracture-vug type reservoir body stratum model; establishing finite element functions according to the DC logging response of the equivalent fracture-vug type reservoir body stratum model; carrying out grid discretion and resistivity assignment on the equivalent fracture-vug type reservoir body stratum model based on finite element theory; and assembling and solving finite element conductance array and obtaining depth lateral logging value. The method solves the problem of trans-scale grid dissection of a fracture-vug type reservoir body, effectively carves the boundary of the fracture-vug type reservoir body, scientifically simulates the values of fracture-vug type reservoir body electric logging, and provides theoretical basis for identifying and evaluating electric logging of carbonatite and volcanic fracture-vug type reservoir body according to electric logging data.

The invention discloses a wheat ear detection and counting method based on a deep learning point supervision idea, and the method comprises the steps: carrying out the pre-collection of a field wheatear image of a specific region, and obtaining an input image; inputting the input image into a network structure and obtaining output parameters; and obtaining each wheat ear patch and predicting thenumber of wheat ears. According to the network structure, firstly, features are extracted through a down-sampling network, and then the extracted features are subjected to up-sampling through an up-sampling path, so that the size of output is consistent with that of an input image, the probability of each pixel in the output image is obtained, and plaques of wheat ears are obtained based on the probability. The method provided by the invention can effectively overcome noise in a field environment, and realizes rapid detection and accurate counting of wheat ears. The reliability of the method is verified through various pictures of different field environments, different illumination intensities, different growth conditions and different shooting lens distances of two cities.

The invention belongs to the technical field of depth learning and radiotherapy, and relates to a method for segmenting human organs in medical images based on a neural network. The method comprises the following steps: roughly positioning a medical image layer where a target organ to be delineated is located, wherein the target organ to be delineated comprises a plurality of organs; A 3D binary U-shaped network is used to determine the initial and final layers of the target organ to be delineated as a whole. The middle layer of the beginning layer and the end layer is input into a 2D multi-class U-shaped network for fine delineation and segmentation, and the corresponding beginning layer and the end layer of each organ to be delineated are determined. The present invention uses a 3D network to determine a starting layer and an ending layer when multiple organs are integrated, A 2D network is use to solve that data imbalance problem between the organs between the start lay and the endlayer, and by combining the advantages of the existing 3D network and the 2D network, the problem of segmentation of multiple organ with similar HU values in the prior art is solved.

The invention belongs to the field of image processing, and provides a method and a device for image segmentation, a computer readable storage medium and terminal equipment so as to enhance the robustness of fuzzy image segmentation with noise. The method comprises the following steps that: carrying out local gray-level mapping processing on a to-be-segmented image subjected to Gaussian smoothingto obtain a first image; carrying out mean value filtering on the first image to obtain a second image; according to the category uncertainty and the regional stability of a pixel gray level, constructing an energy function E(t), and solving to obtain the corresponding threshold value toptimum when the energy function E(t) is minimum; and taking toptimum as a segmentation threshold value to segment the to-be-segmented image. By use of the technical scheme provided by the invention, on one hand, an influence caused by uneven grey level distribution can be effectively reduced, on the other hand,the segmentation problem of a fuzzy image under different noise levels can be effectively solved, and the method exhibits better robustness on the image with the noise.

The invention discloses a part identification method based on a multilayer random forest. The part identification method comprises the following steps: firstly, establishing an image training set anda test set; performing depth feature extraction; establishing a random forest classifier through training; and finally, constructing a multi-layer random forest for classification and identification.The beneficial effects of the invention are that the part identification method can achieve the recognition of the assembly state of the assembly and the recognition of parts at the same time, and canachieve the effective segmentation and recognition of the parts of the assembly in different assembly states.

The invention relates to the field of medical image processing and computer vision, particularly relates to a three-dimensional tooth model grid segmentation method based on a local attention mechanism. The method comprises the following steps: judging and classifying each triangular mesh in a three-dimensional tooth model by adopting a trained three-dimensional tooth model segmentation network, determining whether the area where each triangular grid is located belongs to gingiva or a certain tooth, and accurately finding the complete area where each tooth is located in the three-dimensional tooth model, and therefore the three-dimensional tooth model is accurately segmented. According to the method, the segmentation accuracy of low-feature-recognition-degree regions such as tooth edges and adjacent teeth can be effectively improved.

The invention discloses an improved small curved arm image segmentation method for sign language images. The method comprises: reading an input sign language image, performing YCbCr color space conversion, and performing skin color clustering and segmentation to obtain a skin color region in the image; smoothing processing is carried out through median filtering, burrs or white points around a skin color area are removed, maximum threshold binaryzation is carried out, and holes of a sign language area are filled with an image area filling algorithm; processing by adopting area operator and centroid positioning to obtain hand-arm area and method for judging position of cutting line by using transverse distance and the complete segmentation of the hand area is realized through the positioning of the arm area cutting line. According to the method, the transverse distance is used for judging the position of the cutting line for fine segmentation, the cutting line between the arms is positioned, complete segmentation of the hand area is achieved, the problem of segmentation of the arm area is solved, the method is suitable for segmentation of the small bent arm, and the arm removing method has good robustness.

The invention relates to a fingerprint image background segmentation method, comprising the following steps: step 1, coarsely extracting a fingerprint information boundary value according to difference sum of inter-frame image blocks of adjacent fingerprint sequences; step 2, searching for a stable boundary point according to a difference absolute value of the boundary values of two adjacent points; step 3, processing abnormal boundary values according to the difference absolute value of the adjacent boundary points and a deviation position locating method of the adjacent boundary points; step 4, smoothening the boundary values according to a fingerprint boundary deviation direction; step 5, according to left and right boundary values, substituting a fingerprint image background, thereby making the fingerprint image background more cleaner. The fingerprint image background segmentation method provided by the invention is simple, has fast arithmetic speed and can be realized in a universal ARM chip; and the method is more accurate in segmentation for low-quality fingerprint images and the fingerprint images with mutational background, has wide adaptive range, and can be used for the fingerprint images acquired by a sliding type fingerprint sensor, a planar type fingerprint sensor, and so on.

The invention relates to a small sample tooth detection method based on a prototype segmentation network and a storage medium, a network inputs an analog image and a real image, two sample input networks can respectively obtain virtual image features and real image features after CNN feature extraction, the analog virtual image features are combined with a Mask label, and the real image features and the real image features are combined. Then feature clustering is carried out through an NPR module, real image features are divided into two branches, one branch obtains a preliminary segmentation result under feature guidance of the clustering features, the other branch carries out target detection network, a finally detected result and the preliminary segmentation result are fused, and the segmentation and detection results are mutually promoted to realize target detection. And an accurate tooth segmentation result is obtained. According to the method, the problem of small sample tooth segmentation is effectively solved, the respective advantages of target detection and semantic segmentation are utilized, mutual promotion and mutual learning are realized, the problems that tooth segmentation types are wrong and tooth boundaries are inaccurate are solved, and a large amount of time cost and labor cost can be saved.

The invention provides a character segmentation method based on character features, which is applied to the technical field of image processing and comprises the steps of obtaining a to-be-processed image; performing binarization processing on the to-be-processed image to obtain a binarized image; performing feature extraction on the binary image by adopting a basic feature extraction network; forthe extracted features, carrying out feature extraction on the forms of the characters to obtain first features, and carrying out feature extraction on the number of the characters to obtain second features; fusing the first feature and the second feature by adopting a semantic segmentation network to generate a semantic segmentation graph; and determining a segmentation position of the characteraccording to the semantic segmentation graph. In addition, the invention further discloses a character segmentation device based on the character features and a computer storage medium.

The invention belongs to the technical field of scene segmentation in computer vision, and discloses a mirror image segmentation method based on depth perception. The PDNet sequentially comprises a multi-layer feature extractor, a positioning module and a description module, wherein the multi-layer feature extractor acquires context features by using a traditional feature extraction network; the positioning module preliminarily determines the position of the mirror in the image by combining the RGB feature information with the depth feature information; and the description module is used for adjusting and determining the boundary of the mirror by combining depth information on the basis of the RGB feature information of the image. The method is the first method for realizing mirror segmentation in the image by simultaneously using the RGB image and the depth image. According to the method, further testing is carried out, the PDNet segmentation result is still excellent for a mirror with a large area in a complex environment, and the result at the boundary of the mirror is also satisfactory. The method is wider in applicability.