1270results about How to "Accurate segmentation" patented technology

The invention provides a multi-scale feature fusion ultrasonic image semantic segmentation method based on adversarial learning, and the method comprises the following steps: building a multi-scale feature fusion semantic segmentation network model, building an adversarial discrimination network model, carrying out the adversarial training and model parameter learning, and carrying out the automatic segmentation of a breast lesion. The method provided by the invention achieves the prediction of a pixel class through the multi-scale features of input images with different resolutions, improvesthe pixel class label prediction accuracy, employs expanding convolution for replacing partial pooling so as to improve the resolution of a segmented image, enables the segmented image generated by asegmentation network guided by an adversarial discrimination network not to be distinguished from a segmentation label, guarantees the good appearance and spatial continuity of the segmented image, and obtains a more precise high-resolution ultrasonic breast lesion segmented image.

The invention discloses a liver division method based on a CT image, which comprises the following steps: firstly, an MSCTP artery sequence image and a hepatic portal vein sequence image of the abdomen are pretreated, and the outline of the liver is automatically divided to obtain the liver image; secondly, a multi-scale filtering method based on a Hessian matrix is used for strengthening blood vessels, a dividing method such as region enlargement, and the like is used for dividing the hepatic portal vein, and a three-dimensional topology thinning method is used for picking up the central line of the hepatic portal vein; the blood vessels are identified mutually in classification; distance change and Voronoi arithmetic are then used for calculation, and the outline of the liver is used for covering values to obtain division results; the three-dimensional liver division results are finally reconstructed. The system comprises a liver division module, a blood vessel strengthening dividing and a thinning module, a blood vessel classifying module, a liver dividing module and a three-dimensional reconstructing module. The invention can reduce noise as well as avoid the vagueness of the boundary of the liver, effectively and exactly divides the outline of the liver, improves the division quality of the hepatic portal vein tree, and realizes accurate and rapid division of the liver.

The invention belongs to the technical field of the pattern identification and image processing, and relates to a license plate detection and identification method based on the maximum stable extremal region, which comprises the following steps: extracting the maximum stable extremal region (MSER) to obtain text regions of candidate license plates; adopting one effective feature for description of each extremal region, classifying the extremal regions into 'text' regions and 'non-text' regions with a classifier obtained in prior training, and extracting a license plate from original images based on the characteristics of the structure of the license plate; and describing with the characteristics of the shape context, and completing the character identification through template matching. Since the maximum stable extremal region has the affine invariance, high stability and multi-scale features, and the regions are determined only according to the gray value and are not light-sensitive,the license plate detection and identification method with the maximum stable extremal region as the base is applicable to the complex background, and has the advantages of good stability and high identification rate.

The invention discloses a method for splitting a news video program and a method and a system for cataloging news videos. The method for splitting the news video program comprises the following steps of: sequencing detection results according to a time sequence to obtain an event sequence by detecting characteristic information of titles of the news video, a headline, characteristic information of comperes, lens transformation, a mute point of an audio, a switching point, a keynote period sudden change point and the like; briefing the event sequence by adopting a preset symbol set and a production rule, and judging rough positions of start points and end points of news sections in the event sequence; calculating a union posterior probability of a start position of each news section near the rough start position according to the event sequence, selecting the moment with the maximum posterior probability as the accurate start position of each news section, splitting the news video, and thus obtaining the news video sections. According to the method, the adopted algorithm is stable and effective; the structural information in the news video can be summarized effectively; the accurate positions of splitting points of the news sections can be determined; and the news video can be split stably and accurately.

The invention provides an MRI (Magnetic Resonance Imaging) brain tumor localization and intratumoral segmentation method based on a deep cascaded convolution network, which comprises the steps of building a deep cascaded convolution network segmentation model; performing model training and parameter optimization; and carrying out fast localization and intratumoral segmentation on a multi-modal MRIbrain tumor. According to the MRI brain tumor localization and intratumoral segmentation method provided by the invention based on the deep cascaded convolution network, a deep cascaded hybrid neuralnetwork formed by a full convolution neural network and a classified convolution neural network is constructed, the segmentation process is divided into a complete tumor region localization phase andan intratumoral sub-region localization phase, and hierarchical MRI brain tumor fast and accurate localization and intratumoral sub-region segmentation are realized. Firstly, the complete tumor region is localized from an MRI image by adopting a full convolution network method, and then the complete tumor is further divided into an edema region, a non-enhanced tumor region, an enhanced tumor region and a necrosis region by adopting an image classification method, and accurate localization for the multi-modal MRI brain tumor and fast and accurate segmentation for the intratumoral sub-regions are realized.

The invention discloses a brain tumor segmentation method based on a deep neural network and a multi-modal MRI image. The method includes steps: constructing the deep neural network, wherein the deep convolution neural network includes two three-layer convolution layers, a three-layer full connection, and a classification layer, an input layer corresponds to the multi-modal MRI image, and each node of an output layer corresponds to a tumor classification label; performing MRI image preprocessing; training a network model; and testing the model, performing normalization on a to-be-segmented tumor image sequence by employing image blocks of an MRI image sequence and mean values and standard deviations thereof in a training process, inputting the normalized image sequence to the deep neural network with the optimization network connection weight, obtaining node values of the classification layer, and obtaining the tumor classification of a to-be-segmented brain tumor image. According to the method, tumor abstract topological characteristic information in the multi-modal MRI image is mined and extracted by employing the deep neural network, and high segmentation accuracy and high segmentation precision can be guaranteed in brain tumor segmentation of the multi-modal MRI images.

Object segmentation and tracking are improved by including directional information to guide the placement of an active contour (i.e., the elastic curve or ‘snake’) in estimating the object boundary. In estimating an object boundary the active contour deforms from an initial shape to adjust to image features using an energy minimizing function. The function is guided by external constraint forces and image forces to achieve a minimal total energy of the active contour. Both gradient strength and gradient direction of the image are analyzed in minimizing contour energy for an active contour model.

The invention provides a vehicle license plate recognition method, and relates to a method for image recognition. The method comprises the following steps of preprocessing an image; partitioning a vehicle region according to color and texture characteristics; extracting a remarkable factor graph of a vehicle region diagram; extracting candidate vehicle license plates by an Adaboost classifier based on expanded Harr-like characteristic; determining the position of a real vehicle license plate from the candidate vehicle license plates; partitioning the marked vehicle license plate from the corresponding vehicle region original diagram; carrying out character segmentation according to structural characteristic; and carrying out character recognition based on the improved template matching method. With the method provided by the invention, the defects that the application scene of the traditional vehicle license plate recognition method is relatively single, some traditional vehicle license plate recognition methods are only suitably used for single vehicle license plate recognition of the single scene and difficultly used for multiple-vehicle license plate recognition of multiple scenes, and the recognition rate is easily affected by strong light, haze and weak light environments are overcome.

The invention discloses an automatic segmentation method for a point cloud of a facade of a large scene city building. The method comprises steps of (1) performing fusion registration on airborne LiDAR point cloud data and on-vehicle LiDAR point cloud data, (2) extracting airborn LiDAR building point cloud data from the airborne LiDAR point cloud data which goes through registration in the step (1), (3) performing segmentation on the point cloud data of a single building based on the airborne LiDAR point cloud data extracted from the step (2), (4) performing contour tracking on the single building segmented by the step (3), (5) performing simplification and normalization processing on a contour line extracted from the step (4), (6) performing rough segmentation on the point cloud of the facade of the building based on the contour line which goes through the simplification and normalization processing in the step (5), and (7) performing fine segmentation on the building facade point cloud which is roughly segmented in the step (6). The automatic segmentation method of the invention can fast and accurately separate the building facade point cloud from the on-vehicle LiDAR point cloud.

The invention discloses a main carotid artery blood vessel extraction and thickness measuring method based on neck ultrasound images. The method comprises the following specific steps of: reading neck ultrasound three-dimensional body data, and marking a main carotid artery axis based on a main carotid artery branching node; sequentially projecting and segmenting the neck ultrasound three-dimensional data in the three-view drawing direction to obtain two-dimensional cross section, coronal plane and vertical plane sequence images; and preprocessing, dividing and reconstructing the two-dimensional cross section, coronal plane and vertical plane sequence images or counting the thickness of intima-media membrane to obtain the final relevant information of internal and external profiles of the neck ultrasound main carotid artery blood vessel wall and the thickness of the blood vessel wall. By the method, the defects that the calculating complexity in the blood vessel dividing method is high, the thickness of the blood vessel wall cannot be accurately measured and an error is likely to be caused by subjective factors during computer-aided diagnosis are overcome, and the internal and external profiles of the neck ultrasound main carotid artery blood vessel wall and the thickness of the blood vessel wall can be completely, rapidly and accurately obtained. In comparison with the manual dividing method, the method is rapid in operation, and can be used for auxiliary diagnosis and prevention and treatment of neck atherosclerosis and cardiovascular disease.

The invention discloses a lane line segmentation method and apparatus. The method comprises: a lane line image is collected; the lane line image is processed by using a convolution neural network to obtain a coarse segmentation result; and according to the coarse segmentation result and the lane line image, an image segmentation model is constructed to carry out subdivision on the lane line image, thereby determine a lane line area. According to the invention, when the image segmentation model is constructed, information of the whole lane line image is used and segmentation of the image segmentation model becomes precise, so that lane line segmentation precision is improved.

The invention relates to a video abstraction method based on moving target detection, and the method comprises the following steps of: carrying out moving target detection and analysis processing on an initial image sequence frame so as to obtain visual feature information of a moving object in each frame of an original video, carrying out target tracking and merging processing between frames and extracting index information of each moving target of the original video; reordering each moving target in time and space, and generating an abstract video by fusing background images, and recording the index information of each moving target in the abstract video; and establishing index correlation for moving targets of each frame of the abstract video and the moving targets of the original video. The invention further discloses a video abstraction device based on moving target detection. The target tracking synthesis method provided by the invention is simple and effective, and can be used for achieving fast and accurate abstraction and concentration of a surveillance video; and the use of index allows freely convenient browsing of the original video clip to which the moving objects occur during playing. The video abstraction method and system based on moving target detection can be used for generating an abstract video at a fast speed, are high in accuracy, and are easily used.

The invention brings forward a traffic label identification method of vehicle-mounted laser scanning point cloud data. The method comprises the following steps: 1, point cloud real-time preprocessing; 2, point cloud structure feature obtaining; 3, multi-scale Markov random field point cloud clustering; and 4, traffic label identification. The method has the following advantages: "surface-shaped", "linear-shaped" and "scattered" features of point clouds are reinforced, differences between points are enhanced, and while under-segmentation is avoided, reasonable segmentation of the scale of a traffic label component can be rapidly realized; classification and identification of a traffic label can be conveniently realized from point cloud data of deficiency of a part of the traffic label, caused by ground object shielding or ground object self-occlusion; the method can effectively satisfy the requirements for rapid extracting, monitoring and identifying city components at present, can be conveniently promoted to the industry of manned or unmanned navigation and barrier avoiding based on computer vision, helps a driver to perform navigation and decision-making under complex road conditions, and effectively reduces the traffic accident probability.

The invention discloses an image type fire flame identification method. The method comprises the following steps of 1, image capturing; 2, image processing. The image processing comprises the steps of 201, image preprocessing; 202, fire identifying. The fire identifying comprises the steps that indentifying is conducted by the adoption of a prebuilt binary classification model, the binary classification model is a support vector machine model for classifying the flame situation and the non-flame situation, wherein the building process of the binary classification model comprises the steps of I, image information capturing;II, feature extracting; III, training sample acquiring; IV, binary classification model building; IV-1, kernel function selecting; IV-2, classification function determining, optimizing parameter C and parameter D by the adoption of the conjugate gradient method, converting the optimized parameter C and parameter D into gamma and sigma 2; V, binary classification model training. By means of the image type fire flame identification method, steps are simple, operation is simple and convenient, reliability is high, using effect is good, and the problems that reliability is lower, false or missing alarm rate is higher, using effect is poor and the like in an existing video fire detecting system under a complex environment are solved effectively.

The invention belongs to the technical field of intelligent transportation, and relates to a front vehicle detection method based on monocular vision, which comprises the steps of: acquiring a vehicle front road condition original image, capturing a subimage, carrying out histogram equalization on the subimage, extracting a vehicle front road surface average gray threshold; obtaining a gray threshold by adopting an improved OTSU method; calculating a binarization threshold; estimating a binarization result, segmenting and reinforcing a target according to an estimation result; filtering; obtaining a key position line expressing a bottommost line in a vehicle bottom shadow, carrying out line fusion on the key position line, then extracting target information of an image to be used as a target result of a current frame; and matching the target result of the current frame with a target result of the last frame and information of a tracking target result, and carrying out classifying decision on the target result of the current frame according to a matching result, and determining a final detection result of a front vehicle of the current frame. The front vehicle detection method has the characteristics of good instantaneity, high detection accuracy rate and good robustness.

The invention relates to medical image processing and aims to provide a multiple organ segmentation method based on a deep convolutional neural network and a regional competition model. The multiple organ segmentation method based on the deep convolutional neural network and the regional competition model comprises processes of training a three-dimensional convolutional neural network; using the trained three-dimensional convolutional neural network to learn prior probability images of liver, spleen, kidney and background in CTA volume data; determining an initial segmentation region of each tissue according to the prior probability image of each tissue; determining the probability of each pixel point belonging to each of the four tissues in the image; establishing a multiple region segmentation model based on regional competition; solving the model using the convex optimization method; and performing post-processing to obtain the contour of each organ. The invention uses the convolutional neural network to automatically and rapidly detect positions of liver, spleen and kidney at the belly, thereby obtaining the prior probability image of each organ is obtained. Then the invention uses the regional competition model, so that the contours of liver, spleen and kidney can be accurately segmented.

A method and system detect speech activity. An ultrasonic signal is directed at a face of a speaker over time. A Doppler signal of the ultrasonic signal is acquired after reflection by the face. Energy in the Doppler signal is measured over time. The energy over time is compared to a predetermined threshold to detect speech activity of the speaker in a concurrently acquired audio signal.

A method and system for aorta segmentation in a 3D volume, such as a C-arm CT volume is disclosed. The aortic root is detected in the 3D volume using marginal space learning (MSL) based segmentation. The aortic arch is detected in the 3D volume using MSL based segmentation. The ascending aorta is tracked from the aortic root to the aortic arch in the 3D volume, and the descending aorta is tracked from the aortic arch in the 3D volume.

The invention discloses a steel rail surface defect image adaptive segmentation method. The method comprises the following steps of S1, extracting a steel rail region by adopting a row grayscale mean successive summation method; S2, preprocessing a steel rail region image; S3, performing structure region and non-structure region division on the steel rail region image; S4, further distinguishing a defective region and a shadow region by utilizing a non-local feature of the image in the structure region; S5, adaptively building a background image model according to different features in the image; S6, performing image difference; and S7, performing dynamic threshold segmentation. The image is divided into the structure region and the non-structure region by utilizing image local information, the size of a pixel neighborhood window is adaptively adjusted by utilizing non-local information to calculate a mean, the accurate background image model is built, and the image difference and the dynamic threshold setting are performed, so that while a defective part of the image is highlighted, the influence of uneven illumination and steel rail surface reflection property on steel rail surface defect detection is effectively reduced, an ideal image segmentation effect is achieved, and the rail surface detection precision is ensured.

The invention provides a semantic image segmentation method and system based on edge enhancement. The method comprises the steps of preprocessing an input image; establishing an edge enhancement network model, wherein the edge enhancement network model comprises a lightweight edge network and a deep semantic network; inputting the preprocessed image into a lightweight edge network, and adaptivelypaying attention to local edge information of the image by utilizing a spatial attention module; inputting the preprocessed images into a deep semantic network in batches, and optimizing the output ofthe deep network at different stages by using a channel attention module; carrying out cascading dimensionality reduction on the obtained features, fusing feature information of different levels, andadopting a channel attention module to perform optimization; performing normalizing to obtain an image segmentation result predicted by the edge enhancement network model; and calculating cross entropy loss and focus loss of the predicted segmentation image and a given standard segmentation image so as to respectively supervise output of the lightweight edge network and the deep semantic network,and updating model parameters of the edge enhancement network by using a random gradient descent method so as to realize accurate segmentation of an input image.

A programming method of the multi-level flash memory comprises shooting a programming voltage that is increasing upwards stepwise each time into the gate of the multi-level flash memory, and following, shooting a program verify voltage that is decreasing downwards to program a multi-level in the multi-level flash memory and shooting an additional programming voltage into the multi-level flash memory after the last program verify voltage is shot. An erasing method of the multi-level flash memory comprises shooting an erasing voltage that is decreasing downwards stepwise each time into a gate of the multi-level flash memory, and following, shooting a erase verify voltage that is increasing upwards to erase a multi-level in the multi-level flash memory and shooting an additional voltage into the multi-level flash memory after the last erase verify voltage is shot.

The invention provides a face detection method based on a Gaussian model and a minimum mean-square deviation, and relates to a face recognition technology. A face detection method based on the Gaussian model and the minimum mean-square deviation under the premise of a complicated background, a side face, a stopper and a plurality of faces. The method comprises the following steps of: building a YCbCr Gaussian model: building the YCbCr Gaussian model for face skin color distribution according to collected skin color sample data, and performing lighting compensation on the image, wherein in the YCbCr, Y is a brightness component, Cb is a blue chroma component, and Cr is a red chroma; performing skin color segmentation on the image by using the built YCbCr Gaussian model and the minimum mean-square deviation; performing binaryzation on a skin color region, and processing a binary image by opening to eliminate a small bridge and discrete points; rejecting detected non-face regions in the similar skin color or skin color according to future knowledge of a face; and finally marking a face position by using a rectangular frame.

The invention discloses an image enhancement and partition method, which comprises the following steps of I, image enhancement: a processor and an image enhancement method based on fuzzy logic are adopted for performing enhancement processing on an image needing to be processed, and the process is as follows: i, converting from an image domain to a fuzzy domain: mapping the grey value of each pixel point of the image needing to be processed to a fuzzy membership degree of a fuzzy set according to a membership function (shown in the specifications); ii, performing fuzzy enhancement processing by utilizing a fuzzy enhancement operator in the fuzzy domain; iii, converting from the fuzzy domain to the image domain; II, image partition: partitioning a digital image subjected to enhancement processing, i.e. a to-be-partition image, according to an image partition method based on a three-dimensional fuzzy division maximum entropy. The method is simple in steps, reasonable in design, convenient to realize, good in processing effect and high in practical value, and an image enhancement and partition process can be simply, conveniently and quickly completed with high quality.

An object based boundary refinement method for object segmentation in digital images receives an image and a single initial object region of interest and performs refinement zone definition using the initial object regions of interest to generate refinement zones output. A directional edge enhancement is performed using the input image and the refinement zones to generate directional enhanced region of interest output. A radial detection is performed using the input image the refinement zones and the directional enhanced region of interest to generate radial detection mask output. In addition, a final shaping is performed using the radial detection mask having single object region output. A directional edge enhancement method determining pixel specific edge contrast enhancement direction according to the object structure direction near the pixel consists receives an image and refinement zones and performs 1D horizontal distance transform and 1D vertical distance transform using the refinement zones to generate horizontal distance map and vertical distance map outputs. A neighboring direction determination is performed using the horizontal distance map and the vertical distance map to generate neighboring image output. In addition, a directional edge contrast calculation using the neighboring image and input image having directional enhanced region of interest output.

A level set polarization SAR image segmentation method based on polarization characteristic decomposition, belonging to the radar remote sensing technology or the image processing technology. In the invention, a polarization characteristic vector v which is composed of three polarization characteristics: H, alpha and A is obtained by the polarization characteristic decomposition of each pixel point of the original polarization SAR image; the polarization characteristic vectors v of all the pixel points are combined into a polarization characteristic matrix omega so as to convert the segmentation problem of the polarization SAR image from data space to polarization characteristic vector space; and the condition that the characteristic vector definition is suitable for energy functional of the polarization SAR image segmentation is utilized and a level set method is adopted to realize the numerical value solution of partial differential equation, thus realizing the polarization SAR image segmentation. The method provided by the invention takes full use of the polarization information of the polarization SAR image; therefore, the image edge obtained by segmentation is relatively complete so that the local characteristic is maintained better, the robustness for noise is stronger, the stability of the arithmetic is higher and the segmentation result is accurate; and the invention reduces the complexity of data and can effectively improve the image segmentation speed.

The invention relates to a method of quickly segmenting a moving target in a non-restrictive scene based on a full convolution network, which belongs to the technical field of video object segmentation. The method comprises steps: firstly, framing is carried out on the video, and a result after framing is used for making a Ground Truth set S for a sample image; a full convolution neural network trained through a PASCAL VOC standard library is adopted to predict a target in each frame of the video, a deep feature estimator for an image foreground target is acquired, target maximum intra-class likelihood mapping information in all frames is obtained hereby, and initial prediction on the foreground and the background in the video frames is realized; and then, through a Markov random field, deep feature estimators for the foreground and the background are refined, and thus, segmentation on the video foreground moving target in the non-restrictive scene video can be realized. The information of the moving target can be effectively acquired, high-efficiency and accurate segmentation on the moving target can be realized, and the analysis precision of the video foreground-background information is improved.

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

An active matrix electro-wetting on dielectric (AM-EWOD) device includes a plurality of array elements arranged in an array, each of the array elements including array element circuitry, an element electrode, and a reference electrode. The array element circuitry includes an actuation circuit configured to apply actuation voltages to the electrodes, and an impedance sensor circuit configured to sense impedance at the array element electrode to determine a droplet property at the array element. The impedance sensor circuit is operated by perturbing a potential applied to the reference electrode. The AM-EWOD device includes a common row addressing line. The impedance sensor circuit further is operated by supplying voltage signals over the common addressing line to effect both a reset operation and an operation for selecting a row in the array to be sensed. The circuitry isolates the array element from the actuation voltage during operating the impedance sensor circuit.

The invention relates to a pavement crack image detection method. The method comprises the steps of: carrying out graying and filtering processing on a collected pavement image, constructing a pulse coupling neural network PCNN model, utilizing a genetic algorithm to rapidly find advantages of an optimal solution in a non-linear manner in a solution space so as to optimize important parameters of the model, and rapidly and accurately segmenting cracks and a background in the image; then according to the characteristics of the image after the segmentation, carrying out connected domain detection on the whole image, and filtering out the interference of noise and background textures; and finally, extracting a crack skeleton, calculating the maximum widths of the cracks along the normal line of the skeleton, and making marks in the original image. According to the invention, the digital image processing technology is adopted, the genetic algorithm is utilized to optimize the parameters of the PCNN model, optimization searching is accelerated, the iteration times f the PCNN are reduced, and the iteration is more liable to come to convergence, the interference resistance of the segmentation effect is relatively high, and the segmentation is more accurate; in addition, the modes of connected domain rectangularity, circularity filtering and irregular noise filtering are utilized to filter out a large number of irregular patches, and convenience is provided for the crack detection.

The invention discloses a method and a system for gesture recognition and man-machine interaction based on a 2D video sequence. The method comprises the steps of S1, acquiring a monocular 2D video frame sequence image, and extracting moving foreground in the image; S2, detecting a human hand in the moving foreground, and constructing a joint feature model of the human hand; S3, predicting a location area where a human hand target emerges, searching and positioning the human hand target in the location area by using the joint feature model of the human hand, and acquiring the location of the human hand in the current frame; S4, judging the type of the current operating mode according to the location of the human hand in the current frame; and S5, tracking the human hand, recognizing the posture and the gesture of the human hand in the current operating mode, converting the posture and the gesture into a corresponding instruction so as to realize man-machine interaction. The man-machine interaction method disclosed by the invention not only can realize target hand selection at a complicated background, but also can realize high-accuracy and high-stability tracking for the human hand.