169 results about "Contour segmentation" patented technology

A system for measuring a pulse and respiratory rate from passive thermal video includes contour segmentation and tracking, clustering of informative pixels of interests, and robust dominant frequency component estimation. Contour segmentation is used to locate a blood vessel region to measure, after which all pixels in the nearby region are aligned across frames based on the segmentation's position, and scale in each frame. Spatial filtering is then performed to remove noise not related to heart beat and then non-linear filtering is performed on the temporal signal corresponding to each aligned pixel. The signal spectrum of each pixel is then feed to a clustering algorithm for outlier removal. Pixels in the largest cluster are then used to vote for the dominant frequency, and the median of the dominant frequency is output as the pulse rate.

A method for automatic left ventricle segmentation of cine short-axis magnetic resonance (MR) images that does not require manually drawn initial contours, trained statistical shape models, or gray-level appearance models is provided. More specifically, the method employs a roundness metric to automatically locate the left ventricle. Epicardial contour segmentation is simplified by mapping the pixels from Cartesian to approximately polar coordinates. Furthermore, region growing is utilized by distributing seed points around the endocardial contour to find the LV myocardium and, thus, the epicardial contour. This is a robust technique for images where the epicardial edge has poor contrast. A fast Fourier transform (FFT) is utilized to smooth both the determined endocardial and epicardial contours. In addition to determining endocardial and epicardial contours, the method also determines the contours of papillary muscles and trabeculations.

The invention discloses a digital human serial section image segmentation method based on geometric active contour target tracking, comprising the following steps: 1) inputting a first layer digital human serial section image which includes a target to be segmented; 2) initializing the target contour by intelligent scissors; 3) generating a target contour matching score image; 4) smoothening the target contour matching score image generated in step 3); 5) using a self-adopting geometric active contour model tracking to obtain the target contour of current layer; and 6) storing the target contour segmentation result of current layer, simultaneously taking the result as an initial target contour of next layer and returning to step 3) to continue the target tracking to obtain subsequent target contour. The digital human serial section image segmentation method can accurately and robustly segment the target contour of the digital human serial section image and simultaneously has the advantages of high automation level and processing of topology change of the target contour.

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 multi-target picture segmentation based on a level set, and the multi-target picture segmentation based on the level set is applied in the technical field of the image analysis. The method includes the following steps: firstly, drawing one closed curve or more than one closed curves on to-be-segmented images as an initial outline; secondly, utilizing an initiative outline model based on a region, carrying out an iteration revolution on the initiative outline, and obtaining a profile curve of a target. The initiative outline model based on the region fully takes partial gray level information of the image into account. Therefore, the multi-target picture segmentation based on the level set is capable of segmenting images with uneven gray levels, extending the model into multiple orders, and segmenting over the multi-target images. Compared with the prior art, the multi-target picture segmentation based on the level set has the advantages of being insensitive in initialization, quick in calculation speed, strong in anti-noise ability and accurate in segmentation results. Meanwhile, the multi-target picture segmentation based on the level set is capable of segmenting medical images provided with a plurality of different targets.

The invention relates to a multimodal carotid MRI registration method based on physical alignment and contour matching. The method comprises steps that 1), image space physical coordinates are calculated; 2), body pixel spacing conversion image formats are unified; 3), the alignment relationship of multi-modal image layers is determined; 4), inner wall segmentation of multi-modal 2D images of eachsequence is carried out, a sequence of MRI images is sequentially selected, segmentation start and termination layers of the sequence images are determined, the ROI where a carotid is located in thestart layer image is selected, and the segmentation method is utilized to realize continuous automatic segmentation of the entire sequence; and 5), 3D registration based on the inner wall contour is carried out, and the clear inner wall contour segmentation result is utilized to carry out three-dimensional continuous registration. The method is advantaged in that carotid multimodal sequence imagesin various types are effective, alignment registration among pixels is excellently realized, and doctors are facilitated to carry out subsequent diagnosis and treatment determination and patch composition analysis.

The invention relates to a target recognition method based on the spatial relationship of contour segments, which comprises the following steps: (1) establishing a database of multi-class target images; (2) extracting the peripheral contour of the target and generating a contour point set; (3) describing the shapes of the two contours according to the contextual shape features of the contour pointset, and obtaining the rough matching results of the targets according to the similarity measurement results of the two contours; (4) constructing the spatial relationship between the images in the library and the occluded images to be recognized according to the skeleton of the whole image and the centroid of the outline of the discrete image; (5) establishing the constraint standard of the characteristic parameters of the spatial relationship, and measure the similarity according to the constraint standard; in the case of complex occlusion, more feature information is provided for recognition in the process of target recognition. When the target is occluded, the target is often divided into several parts, and the outer contour no longer has integrity. In this case, considering the spatial relationship between different contour segments, the recognition rate of occluded target is increased.

The invention provides a video image motion target extracting method based on pattern detection and color segmentation. The method comprises the steps of: firstly preprocessing video signals to obtain an RGB (red, green and blue) image, performing k-means image clustering segmentation to the RGB image, and recording color classification number of each pixel point; secondly, performing LBP (local binary pattern) detection according to the gray value of the current frame and the background frame, calculating to obtain the pixel points representing a motion target, and creating a mapping relation between the pixel points and macro blocks to obtain the macro block level motion target; then, according to the color classification number and pixels of the motion target, performing overlapping detection to obtain the initial motion target; and at last, merging the motion targets obtained from the LBP detection and the color segmentation, and filtering to obtain the final motion target. The method has the advantages that the advantages of accurate location in the pattern detection and robustness of illumination variation are kept; and the problem that the motion target is fused in a background by slow movement or in-situ movement can be effectively solved.

The invention discloses a pedestrian individual segmenting method and device. The method includes: utilizing a thick-granularity human shape profile segmenting model which is well trained in advance to perform pedestrian segmenting on a to-be-processed image to acquire a blocked segmenting result which includes multiple blocks marked as background and foreground, wherein the blocks marked as the background in the to-be-processed image do not include pedestrian subject, and the blocks marked as the foreground include part of images of the pedestrian subject; removing a background image in a part corresponding to the blocked pedestrian segmenting result in the to-be-processed image to acquire a thick-granularity segmenting image; inputting the thick-granularity segmenting image into a thin-granularity human shape profile segmenting model which is trained well in advance; outputting an individual pedestrian segmenting result by the thin-granularity human shape profile segmenting model which is well trained in advance, wherein each of the thick-granularity human shape profile segmenting model and the thin-granularity human shape profile segmenting model is acquired by training through a full-convolution neural network.

The invention discloses a calibration method for the steel rail profile measured by adopting a laser displacement technology. The calibration method comprises the steps of segmenting the steel rail profile, and calculating an index value; acquiring absolute affine invariants; acquiring a local affine non-deformation description feature vector set in an offline state and a local affine non-deformation description feature vector set in an operating state, matching feature vectors and acquiring a matching point set; calculating a transformation relational expression of profile data in the offline state and a transformation relational expression of profile data in the operating state; acquiring an affine transmission matrix; acquiring an affine transmission matrix set for each pair of matching points and completing data processing; and performing calibration on the steel rail profile according to the affine transmission matrix, the translation amount and the profile data transformation relational expressions. The invention innovatively proposes a new method for extracting, describing and matching the steel rail profile local affine invariants, puts forwards a quick iterative closest point algorithm to precisely adjust affine transformation parameters, and solves a problem of profile deformation in steel rail profile measurement performed by adopting the laser displacement technology, so that accurate correction is performed on the steel rail profile. In addition, the method is high in calculation speed and high in accuracy.

The invention relates to a real-time pose tracking method based on a target three-dimensional model. According to the method, based on a three-dimensional model of a target, contour features are utilized to establish a contour component model, and a contour is divided into a plurality of local contour sections according to the local contour recognition degree; For each contour segment, two dimensional-three dimensional correspondence is established in a gradient domain based on a phase matching criterion; a wrong matching pair is established by utilizing a geometrical relationship among all parts of the rigid body target; based on the established two dimensional-three dimensional matching pair, a pose parameter is solved through an RPNP algorithm, the whole process is iterated until convergence, and target pose tracking in the sequence image is realized. The algorithm is simple and clear in step and easy to apply, and drift-free tracking in the whole process can be achieved due to introduction of the three-dimensional model of the target.

The invention provides a method for automatically partitioning a complicated section zone in an additive manufacturing technology and belongs to the field of additive manufacturing. The method comprises the steps that a complicated cross-section outline with holes is divided into a plurality of simple inner-hole-free sub-zones. Continuous printing in each sub-zone can be completed without cutter lifting. After zone partitioning, the zones are independently processed, and the cutter lifting only occurs in the zone connecting process. Therefore, the phenomenon that a simple parallel scanning route needs to step over the inner hole zones of non-processing routes for multiple times is avoided. The method aims at the complicated section zone having holes and other shape characteristics, the problem of section zone partitioning optimization due to the fact that a target is the minimum idle stroke of a mechanism is solved, and optimization of scanning route tracks is achieved. The section zone partitioning method adopts a partitioning point distinguishing algorithm, a zone partitioning cross point algorithm and a zoned outline information storage algorithm. The phenomenon that the simple parallel scanning route needs to step over the inner hole zones of non-processing routes for multiple times can be avoided, and accordingly the additive manufacturing efficiency can be improved.

The invention relates to the technical field of three-dimensional image generation, and specifically relates to a three-dimensional modeling method of strawberry plants. The method comprises the following steps: calculating a central point of vector field convolution, selecting a multi-objective initial contour control point, and applying an active contour model of a parameter to intensity image over-segmentation; and adopting a labeling method to convert a segmented contour into a point cloud set, taking the point cloud set as a unit to construct a plane fitting selection mechanism, and finally finishing modeling and display of a three-dimensional model of the elevated in-situ strawberry, and carrying out an exploratory research for the construction of a local vision scene for an agricultural intelligent robot. The method of the invention can perform three-dimensional modeling on in-situ plants, which effectively overcomes the limitation that the traditional modeling method cannot build a three-dimensional model of a whole plant in a greenhouse environment, and improves the utilization ratio of collected data.

A system and method for active contour segmentation. An image of a first and at least a second structure, a first position for the first structure, and at least a second position for the at least second structure are received, the first and the at least second positions are set as a first and an at least second initial contour, and the first and the at least second initial contours are concurrently and iteratively deformed to respectively expand into a first and at least a second expanded contour matching a shape of the first and the at least second structure by applying constraints to each point of the first and the at least second initial contour, comprising a constraint for preventing the first and at least one of the at least second initial contour from intersecting one another upon being deformed, and the constraints are updated after each iteration.

The invention provides a hybrid level set based three-dimensional tooth modeling method. The method comprises the following multiple steps: 1) selecting an initial slice and initializing a level set; 2) segmenting a tooth root slice image by utilizing a single-phase hybrid level set model constructed by combining priori shape constraint energy, Flux model based edge energy and priori gray based local region energy; 3) segmenting a tooth crown layer slice image by a dual-phase hybrid level set model in combination with regional competition constraint; and 4) reconstructing a contour segmentation based three-dimensional tooth model. According to the method, the hybrid level set model can effectively solve the problems of inaccuracy for edge positioning, incapability of processing an image, uneven gray level and the like in a conventional segmentation model, and an independent three-dimensional model of each tooth can be quickly and accurately constructed, so that a solid foundation is laid for making oral repair plans, biomechanical analysis and the like.

The invention discloses a method and a system for building an image database. The method comprises the following steps of utilizing keywords to automatically download images from the Internet, carrying out object detection on the downloaded images, carrying out preliminary filtering, carrying out fine contour segmentation on detected objects to obtain segmentation pixels, and filtering the obtained segmentation pixels based on contour match to form the image database. The database can generally achieve 80-90 percent of tagging accuracy, and a user can quickly carry out content-based retrieval on the image database in modes such as draft drawing or framework adjustment so as to quickly find the needed image synthesis and creating resources.

The invention belongs to the field of automatic recognition and classification of underwater targets, in particular to an automatic classification method of side scan sonar image targets based on transfer learning and depth learning. The method includes acquiring a conventional optical image data set with segmentation annotations; carrying out the contour segmentation by using the annotated imagescorresponding to each image in the data set; selecting a convolution neural network structure for training to obtain the source domain target classification network; fozening the parameters of the front part of the fully trained source domain target classification network, and setting the parameters of the back part of the classification network to the trainable state; continuing to train the setclassification network using the training set; when the training is complete, using the verification set to evaluate the performance of the classification network. The method uses the transfer learning method to transfer the convolution neural network trained with non-side scan sonar images, and pre-processes the source domain data set according to the similarity principle, so as to improve the transfer learning efficiency and prevent the negative transfer phenomenon.

The invention discloses an end-to-end colorectal polyp segmentation method based on effective learning. According to the method, a network model combining a full convolutional neural network with a fully connected conditional random field recurrent neural network is proposed and designed by using a deep learning technology and is used for colorectal polyp segmentation. For the phenomenon that polyp region feature training is insufficient and unrelated organization learning is excessive in the network training process, an effective learning loss function is introduced, and the loss function refers to difficult case mining and boundary perception ideas. The loss function refers to an objective function of a difficult case mining thought, sample training errors are weighted for the problem ofnon-uniform proportion of a colorectal polyp foreground region and a background region, the attention degree of a difficult case is improved, and meanwhile, the attention degree of boundary pixels isimproved in combination with boundary factors to improve the contour segmentation precision.

The invention provides an improved object-oriented high-resolution remote sensing image classification method. The method comprises the following steps: firstly, carrying out initial segmentation on an image by using an SLIC superpixel segmentation algorithm; enabling an initial segmentation result to generate over-segmentation; subjecting the initial segmented image to region fusion to solve anover-segmentation problembut a local under-segmentation phenomenon occurs in the image subjected to region fusion, so that secondary segmentation is carried out on the local under-segmentation region toobtain a new segmented image, the over-segmentation and under-segmentation phenomena in the segmentation process can be solved, and the image segmentation and image classification precision is improved.

A depth information static gesture segmentation method includes the steps of converting a depth image into a depth grey-scale image of a same size, determining the gray scale of a gesture area in the depth grey-scale image, converting the depth grey-scale image into a binary image, conducting smooth processing for the binary image to obtain a mask image, determining a brightness component image, and segmenting the gesture area. The gesture area image segmented is accurate, and no segmentation problem exists. The influence of factors including non-uniform illumination, race difference, other human body parts and similar color background on gesture segmentation is prevented. The method is simple and rapid, and provides technical foundation for man-machine interaction works including gesture identification, control and medical surgeries.

The invention provides a large-scale train shift fault detection method and system based on deep learning. The detection method comprises the steps that of collecting and transmitting a train part shifting image is collected and transmitted into a deep learning framework to train a train part classification model, and obtaining a part shear map is obtained; marking the shear map and transmitting the shear map into a deep learning framework to train a component contour segmentation model; collecting a to-be-detected image, transmitting the to-be-detected image into the train part classification model for prediction, automatically detecting parts according to classification and positioning, and shearing; transmitting the shear map into a component contour segmentation model to segment inner and outer contours of a component, and obtaining relative position information; and setting a threshold value of the relative position information, carrying out logic judgment according to the inner and outer contour relative position information, and judging whether the train part has a displacement fault. Compared with the prior art, automatic detection of mass train part data is effectively achieved, the detection efficiency and accuracy are improved, and the manpower and material resource cost is reduced.

The invention discloses a CT image computer-aided diagnosis system and method for the liver disease based on data mining. The system comprises an input module, a texture feature extraction module, a classification diagnosis identification module and an output module. The method adopts the system and comprises the following steps of: segmenting and extracting the contour of an abdominal CT image inadvance, marking the normal liver CT, the liver cyst or the liver cancer, and importing the preprocessed image into the system; analyzing the image texture of the imported image to obtain the 13-dimensional gray-scale co-occurrence texture feature in the CT image by the image texture feature extraction module; according to a database, establishing a diagnosis model, and substituting the input liver CT image texture feature into a diagnosis model classifier for processing to obtain a diagnosis result and accuracy, wherein the database is stored in the classification diagnosis module and is composed of liver cancer and liver cyst image texture features and normal liver image texture features.

The invention is suitable for the field of the medical image technology, and provides a magnetic resonance image-based bone segmentation method and a system thereof. The method comprises the following steps: calculating step: calculating the Hessian matrix of each pixel and a characteristic value thereof in a first magnetic resonance image; extracting step: extracting the flake structure of the first magnetic resonance image and performing gray level enhancement on the flake structure of the first magnetic resonance image to obtain a second magnetic resonance image; optimization step: performing threshold connected calculation on the second magnetic resonance image, extracting a bone contour of the second magnetic resonance image, and carrying out smooth optimization to obtain a bone contour segmentation result of the second magnetic resonance image. Therefore by applying the magnetic resonance image-based bone segmentation method and the system thereof, a bone needing to be segmented is completely and independently segmented from the magnetic resonance image.

The invention discloses a 3D vertebra CT image active contour segmentation method fusing weighted random forest, and relates to the field of medical image processing. For the problem of sensitivity ofa vertebra CT image segmentation method to an initial contour, a method for automatically locating a vertebra and segmenting a vertebra CT image is proposed. The method comprises the steps of firstly, proposing a weighted random regression and classification forest algorithm to determine a vertebra center; secondly, putting an initial contour ball of active contour segmentation in the vertebra center, and segmenting out the vertebra in the image by adopting a 3D active contour segmentation method in combination with an energy function; and finally, performing combination output on trained models to obtain a complete vertebra CT image segmentation model. A spinal CT segmentation model proposed in the method can automatically locate the vertebra center and can perform automatic 3D segmentation on the vertebra, so that the spinal CT image segmentation steps and processes are simplified.