1394results about How to "Improve Segmentation Accuracy" patented technology

The invention discloses a fundus image retinal vessel segmentation method and a fundus image retinal vessel segmentation system based on deep learning. The fundus image retinal vessel segmentation method comprises the steps of performing data amplification on a training set, enhancing an image, training a convolutional neural network by using the training set, segmenting the image by using a convolutional neural network segmentation model to obtain a segmentation result, training a random forest classifier by using features of the convolutional neural network, extracting a last layer of convolutional layer output from the convolutional neural network, using the convolutional layer output as input of the random forest classifier for pixel classification to obtain another segmentation result, and fusing the two segmentation results to obtain a final segmentation image. Compared with the traditional vessel segmentation method, the fundus image retinal vessel segmentation method uses the deep convolutional neural network for feature extraction, the extracted features are more sufficient, and the segmentation precision and efficiency are higher.

The invention provides an automatic segmentation method for an MRI (Magnetic Resonance Imaging) image brain tumor based on a full convolutional network. The method comprises multi-mode MRI image preprocessing of the brain tumor, construction of a full convolutional network model, network training and parameter optimization as well as automatic segmentation of a brain tumor image, specifically, the segmentation of the MRI image brain tumor is converted into a pixel-level semantic annotation problem and differential information emphasizing different modes of MRI, two-dimensional whole slices of four modes FLAIR, T1, T1c and T2 are synthesized into a four-channel input image, the convolutional layer and the pooling layer of the trained convolutional neural network are base feature layers, three convolutional layers equal to a full connection layer are added behind the base feature layers to form a middle layer, the middle layer outputs rough segmentation images corresponding to semantic segmentation types in quantity, and a de-convolutional network is added behind the middle layer and used for interpolating the rough segmentation images to obtain a fine segmentation image having the same size as the original image. The method does not need manual intervention, effectively improves the segmentation precision and efficiency, and shortens the training time.

The invention discloses a computerized auxiliary diagnosing method for malignant tumor in early stage based on deep learning algorithm. According to the invention, a deep convolution neural network is used to extract the high-level characteristics of an image layer by layer so that the obtained characteristic dimensions become smaller and smaller. To realize the dimension matching between an outputted probability predicting image and a pre-segmented image, the method uses the reverse convolution network to expand the dimensions of the characteristic image so as to obtain a dimension-consistent probability predicting image. The network generated probability predicting image extracts the binarization of the predicting image through the training of a softmax classifier to finally obtain the segmentation result of the prostate tissue. To some extent, the method effectively increases the accuracy and efficiency for doctors to diagnose prostate diseases.

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.

The invention discloses a segmentation method for an image with deep image information, which is high in segmentation accuracy, and capable of still achieving a good segmentation effect under the condition of a very similar front background. The segmentation method comprises the steps of: (1) obtaining the image with deep image information via Kinect; (2) performing probability modelling on the colour information of the front background and the deep image information; (3) performing parameter estimation on the model by EM (Expectation-Maximization) algorithm; and (4) performing segmentation after the first image segmentation on the image by adopting an image segmentation algorithm, wherein an energy function is formula shown in the abstract, and according to the energy function, the smallest segmentation is evaluated by a maximum flow algorithm, so as to obtain the final segmentation object.

The invention discloses a generative adversarial network-based pixel-level portrait cutout method and solves the problem that massive data sets with huge making costs are needed to train and optimizea network in the field of machine cutout. The method comprises the steps of presetting a generative network and a judgment network of an adversarial learning mode, wherein the generative network is adeep neural network with a jump connection; inputting a real image containing a portrait to the generative network for outputting a person and scene segmentation image; inputting first and second image pairs to the judgment network for outputting a judgment probability, and determining loss functions of the generative network and the judgment network; according to minimization of the values of theloss functions of the two networks, adjusting configuration parameters of the two networks to finish training of the generative network; and inputting a test image to the trained generative network for generating the person and scene segmentation image, randomizing the generated image, and finally inputting a probability matrix to a conditional random field for further optimization. According tothe method, a training image quantity is reduced in batches; and the efficiency and the segmentation precision are improved.

The invention discloses a network text segmenting method based on the genetic algorithm, used for segmenting short network texts. The method comprises the following steps of: evaluating a Latent Dirichlet allocation (LDA) model corresponding to a corpus by using a Gibbs sampling method, inferring latent topic information using the model, representing texts by using the latent topic information; then transforming a text-segmenting process into a multi-target optimum process by using a parallel genetic algorithm, and calculating the coherency of segmented units, the divergence among the segmented units and fitness functions by using deeper semantic information; and carrying out the genetic iteration of the text segmenting process, and determining whether the segmenting process terminates based on the similarity among multi-iteration results or the upper limit of iterations to obtain the global optimal solution for segmenting the texts. Therefore, the invention improves the accuracy for segmenting the short network texts.

The invention discloses a deep learning-based remote sensing image semantic segmentation method. The method comprises the steps of assigning an RGB value and a gray value to each target species, obtaining an original remote sensing image, selecting a target species to color and gray the target species, imparting a gray value to the target species to obtain a label image, and subjecting the original remote sensing image to data enhancement and edge extraction to obtain an edge-extracted image; training a full convolution neural network by adopting the original remote sensing image and the imagetraining sample of the edge-extracted image to obtain an optimum semantic segmentation network model, and inputting a to-be-tested remote sensing image into the optimum semantic segmentation networkmodel to obtain a semantic segmentation result image; coloring the semantic segmentation result image to obtain a final semantic segmentation result image, and obtaining a species target according toRGB values in the final semantic segmentation result image. According to the method, the semantic segmentation results of remote sensing images are high in accuracy, and the method is wide in applicability.

The invention discloses a multimedia video object processing method which comprises the steps as follows: (1) carrying out scene dividing on an MPEG video based on macro block information; (2) pre-reading a video needed to be jointed, obtaining various information and searching a proper joint scene; (3) searching the inlet point and the outlet point of the joint and carrying out regulation on various information of the accessed video; (4) selecting a proper audio joint point to realize audio-video seamless joint; (5) setting a video buffer area and unifying the code rate of the video to be jointed; (6) carrying out coarse extraction on a moving object in the video in a time domain; (7) carrying out watershed processing on the coarse extraction result, carrying out space region merging and leading to a accurate segmentation object. The .invention is characterized by simple and high-efficiency algorithm, low system resource consumption and fast speed and high accuracy of processing.

The invention relates to a deep convolutional neural network-based traditional Chinese medicine tongue image automatic segmentation method and belongs to the computer vision field and traditional Chinese medicine tongue diagnosis field. According to the method of the invention, a convolutional neural network structure is designed; collected sample data are adopted to train the network, so that a network model can be obtained; and the model is adopted to automatically segment a traditional Chinese medicine tongue image. The method includes an offline training phase and an online segmentation phase. The method can be applied to both closed type and open tongue image acquisition environments and can effectively improve the accuracy and robustness of the automatic segmentation of the traditional Chinese medicine tongue image. The method of the present invention specifically relates to deep learning, semantic segmentation, image processing and other technologies.

The invention provides a cloud atlas segmentation method based on a FCN and a CNN, and belongs to the field of image segmentation of computer vision. The method is characterized in that firstly, a near neighborhood of each pixel point in a cloud atlas is correspondingly clustered through an ultra-pixel, the cloud atlas is input to fully convolutional networks FCN32s and FCN8s with different step lengths, and pre-segmentation results of the cloud atlas are realized; in a FCN32s result map, a black area is a part of a non-cloud area in the cloud atlas, and in a FCN8s result map, a white area is a part of a cloud area in the cloud atlas; and remaining uncertain areas are grey areas which need to be determined through a deep convolutional neural network (CNN), key pixels in an ultra-pixel area need to be selected to represent characterizes of the ultra-pixel area, and whether the characteristics of the pixels are cloud or non-cloud is determined through the CNN network. According to the method, the precision is equivalent to that of MR-CNN and SP-CNN, and the speed is increased by 880 times compared with MR-CNN and increased by 1.657 times compared with SP-CNN.

The invention discloses a remote sensing image segmentation method based on region clustering, belonging to the field of remote sensing image comprehensive utilization. The method comprises the following steps: carrying out region pre-segmentation by a MeanShift algorithm to remove noise and perform initial cluster on image elements; carrying out fuzzy clustering on images which are subject to the pre-segmentation by a fuzzy C-means algorithm (FCM), and initially inducing and identifying characteristics of each image object to obtain the probability that each object affiliates to some a category so as to constitute a land category probability space of the remote sensing images, thereby providing a basis of object combination for further region segmentation; and performing region segmentation in the probability space of clustering images, classifying image elements which are close in the probability space and similar in the category as the same objects by region labels. In the method of the invention, two defects in the existing segmentation method are overcome, the remote sensing images can be effectively and accurately segmented, segmentation tasks of the remote sensing images can be finished by batch by integration, and data support can be preferably provided for extraction of land information from the remote sensing images.

The invention discloses a multiresolution CNN-based millimeter-wave radar cloud map segmentation method, and belongs to the field of image segmentation. The method comprises the following steps of: obtaining a high-temporal-spatial resolution cloud evolution map of a horizontal / vertical structure by utilizing millimeter wave cloud radar, wherein the information of a context of the cloud map is fully utilized; learning local and global features of the cloud map through respectively inputting three image areas with different resolutions into three CNN networks configured with same parameters; classifying the learned features into cloud features and non-cloud features through classifiers of the neural networks, so as to realize the segmentation of the cloud map; and finally combining the segmentation results of the three networks so as to ensure that the segmentation correctness of the cloud map is up to 99.67%.

The invention discloses a deep residual network-based semantic mammary gland molybdenum target image lump segmentation method. The method comprises the following steps of: labelling pixel categories of lumps and normal tissues corresponding to a collected mammary gland molybdenum target image so as to generate label images, and dividing the mammary gland molybdenum target image and the corresponding label images into training samples and test samples; preprocessing the training samples to form a training data set; constructing a deep residual network, and training the network by utilizing thetraining data set, so as to obtain a deep residual network training model; after a to-be-segmented mammary gland molybdenum target image lump is preprocessed, carrying out binary classification and post-processing on a pixel of the to-be-segmented mammary gland molybdenum target image by utilizing the deep residual network training model, and outputting lump segmentation image to realize semanticsegmentation of the mammary gland molybdenum target image lump. The method is capable of effectively improving the automatic and intelligent levels of mammary gland molybdenum target image lump segmentation, and can be applied to the technical field of assisting radiologists to carry out medical diagnosis.

The invention discloses a retinal vessel segmentation method based on combination of deep learning and a traditional method and relates to the fields of computer vision and mode recognition. According to the method, two grayscale images are both used as training samples of a network, corresponding data amplification, including elastic deformation, smooth filtering, etc., is done against the problem of less retinal image data, and wide applicability of the method is improved. According to the method, an FCN-HNED retinal vessel segmentation deep network is constructed, an autonomous learning process is realized to a great extent through the network, convolutional features of a whole image can be shared, feature redundancy can be reduced, the category of multiple pixels can be recovered from the abstract features, a CLAHE graph and a gauss matched filtering graph of the retinal vessel image are input into the network, an obtained vessel segmentation graph is subjected to weighted average, and therefore a better and more intact retinal vessel segmentation probability graph is obtained. Through the processing mode, the robustness and accuracy of vessel segmentation are improved to a great extent.

The invention discloses a building segmentation and contour line extraction method of ground laser point cloud data. The building segmentation and contour line extraction method comprises the steps of vertically projecting a point cloud, generating a two-dimensional grayscale image, segmenting a building point cloud by adopting an Otsu algorithm, carrying out principal component analysis, conducting best neighborhood calculation, extracting contour lines and the like. The building segmentation and contour line extraction method realizes the segmentation of the building point cloud from the original point clout quickly and accurately, and completes the extraction of the building contour line point cloud in a full-automatic manner.

Systems, methods, and ultrasound transceivers equipped and configured to execute harmonic analysis and extract harmonic information related to a targeted organ of a subject are described. The methods utilize neural network algorithms to establish improved segmentation accuracy of the targeted organ or structures within a region-of-interest. The neural network algorithms, refined for detection of the bladder and to ascertain the presence or absence of a uterus, is optimally applied to better segment and thus confer the capability to optimize measurement of bladder geometry, area, and volumes.

The invention discloses an image partition method of synthetic aperture radars (SAR) which is based on a level set and relates to the application technology of radar remote sensing. The method comprises the following procedures: an SAR echoed signal is acquired by a receiver and a hybrid probability model of an SAR image is computed; a boundary detection operator is computed according to the hybrid probability model; an energy functional based on a boundary information is acquired by combining a geodesic active contour model with the boundary detection operator; the energy functional based on a region information is computed and a partition model is defined as the weighted sum of the energy functional which are based on the region information and the boundary information; the partition model is minimized by a variation method, so as to acquire the partition result of the SAR image. As the invention uses the level set method for transforming curve movement into curved surface movement, even if the target boundary is disunited or merged in the image partition, the topology structure of the curved surfaces does not change, and simultaneously the invention does not need a noise preprocessing procedure, thus improving the precision and the applicability of the SAR image partition.

The invention discloses a CT image liver segmentation method and system based on characteristic learning, being able to effectively improve the segmentation precision of the liver in a CT image. The method comprises the steps: S101: reading a training image set and an image to be segmented, wherein the training images in the training image set and the image to be segmented are the CT images of a belly; S102: extracting the Haar characteristic of the training images and the image to be segmented, the local binary pattern characteristic, the directional gradient histogram and the co-occurrence matrix characteristic; S103: utilizing a principal component analysis method to perform characteristic fusion on all the extracted characteristics so as to acquire more effective characteristics; S104: utilizing a classifier to classify the characteristics of each pixel of the image to be segmented to obtain a liver probability graph; and S105: combining the liver probability graph with the image to be segmented, modifying the graph model weight of a random walk segmentation algorithm to realize segmentation of the liver.

The invention discloses a fundus image optic disc segmentation method and a system for assisting glaucoma screening, and relates to the technical field of image information processing. The fundus image optic disc segmentation method comprises the steps that a plurality of fundus images are collected and preprocessed, and a training image sample set and a verification image sample set are obtained;training of a constructed W-Net-Mcon full convolutional neural network by using the training image sample set to obtain an optimal W-Net-Mcon full convolutional neural networkis carried out; preprocessing the fundus image to be segmented, and inputting the preprocessed fundus image to be segmented into the optimal W-Net-Mcon full convolutional neural network to obtain a prediction target result image; Processing prediction target result graph by utilizing polar coordinate inverse transformation and ellipse fitting to obtain final segmentation result so as to obtain cup-to-disk ratio and finally obtain glaucoma preliminary screening result. According to the method, image semantic information can be effectively extracted in a multi-size mode, fusion of features of different levels, fusion of global features and detail features and encouragement of feature multiplexing are carried out, gradient back propagation is improved, and the image segmentation precision is improved.

The invention provides a liver and tumor segmentation method and system based on multitask deep convolution network. The method comprises the steps of conducting data pre-processing and expanding; establishing the multitask deep convolution network; establishing a supervising layer having inter-task constraint; conducting network training; generating data test result. The invention is advantageousin that the multitask network includes two pathways, which are respectively used for realizing the identification and segmentation of liver and liver tumor; through sharing of the pathway part, the volume of the deep convolution network can be reduced, and the demand of model learning on the training data amount can be reduced; through connecting to a characteristic extracting module relevant torespective task and a corresponding output module after the sharing of the pathways, the multitask output can be realized; the aim of mutual constraint of the liver and liver tumor can be achieved bymeans of a supervising and learning module of multitask geometrical association information; liver and liver tumor can be accurately identified and segmented.

The invention discloses a CT image liver segmentation method and system based on an automatic context model and capable of increasing liver segmentation precision in a CT image. The method comprises steps of: reading a training image set and a to-be-segmented image; extracting the textural feature of each pixel in the images; classifying the feature of each pixel in the to-be-segmented image by using a classifier to obtain an initial liver probability graph; extracting the context feature of each pixel in the images; combining the context features with the textural features and learning a series of classifiers by means of iteration until convergence to obtain a liver probability graph; using the liver probability graph as prior information and adding the liver probability graph a prior constraint condition to a randomly walking objective function as to obtain a random walk model based on context constraint for segmenting the liver; achieving liver segmentation of a three-dimensional CT image layer by layer on a two-dimensional slice of the to-be-segmented image, and interpolation and complement of a liver bound discontinuous area so as to obtain a smooth continuous liver surface.

A method and system for purchased-based segmentation of potential customers employs the use of actual, observed purchases instead of presumptions and correlations to improve the accuracy of segmentation and involves collecting empirical data for a client on actual purchasing behavior of a group of customers and applying statistical modeling techniques to the empirical purchasing behavior data to identify segments or clusters of the customers that exhibit similar purchasing propensity characteristics. Thereafter, the segments or clusters are further differentiated from one another according to other factors having a tendency to directly affect actual purchasing behavior of the customers within the segments or clusters, and potential customers are then identified according to a correlation with the segments or clusters for customized marketing.

The invention relates to a method for changing a plane video to a stereoscopic video, based on real time human-computer conversation, which belongs to the multimedia technical field of computers. The method comprises that: a computer divides a whole video sequence into sub-sequences that have related contents; a user designates any frame of each sub-sequence as a key frame; the computer carries out wave-filtration to all plane video frames to enhance image edge information and sharpen the edge of the foreground objects of the video frame; foreground object segmentation is carried out to the key frames and non-key frames and profile curves and depth maps are extracted until depth map sequences corresponding to all the frames of the original plane video sequence are generated; the smoothed depth map sequences are utilized for rendering and generating multi-visual angle views corresponding to each moment, and the views are synthesized into stereoscopic video frames; the stereoscopic video frames of all the moments are composed into the stereoscopic video sequences. As the method is based on the real time human-computer conversation, the precise depth map of each frame can be obtained, thus well realizing the changing from the plane video to the stereoscopic video and finally obtaining the stereoscopic video with better effect.

The invention discloses an ultrasonic image segmentation method of cascaded total convolution neural network for thyroid nodule, which comprises the following steps of: constructing an ultrasonic image segmentation method based on U-NET, the region of interest (ROI) is segmented from the ultrasound images of thyroid gland based on the simplified total convolution neural network (TCNN); using the VGG19-FCN network as the down-sampling layer to extract the deep features of ROI, so as to realize the automatic semantic segmentation of thyroid nodules. The simplified total convolution neural network comprises a fifth convolution layer for downsampling and a fifth upsampling layer for upsampling. The first five convolutions conv are composed of two 3x3 convolution layers and one pooling layer, each convolution layer uses ReLU as the activation function, and the last five upsampling layers are deconvolution layers. The invention provides a high-precision nodule image for identifying benign and malignant thyroid nodules, thereby playing a better auxiliary role in medical diagnosis.

The invention discloses a framework for improving the effect of semantic segmentation model based on transfer learning, which comprises the following contents: 1) introducing the transfer learning into the semantic segmentation field, so that the fast semantic segmentation network can improve the segmentation effect of the student model through the teacher model; 2) consistency mapping being proposed to measure the contour and texture information of the teacher-student model, and a consistency loss function being constructed to make the fast semantic segmentation better in detail. 3) using a teacher model and a conditional random field (CRF) model to generate auxiliary label for unlabeled data, and adding data to a training set to improve that generalization ability and segmentation effectof the model. The invention improves the accuracy rate of the fast semantic segmentation model without introducing additional model parameter and reducing the model speed.

The invention discloses an image semantic segmentation model generation method and device, equipment and a storage medium. The method comprises the steps of obtaining an image sample set; using the image sample set to train an image semantic segmentation model; wherein the image semantic segmentation model comprises a feature map extraction part and a feature map analysis part, the feature map extraction part comprises a plurality of cascaded hole convolution processing residual error modules, and the feature map analysis part is constructed based on an attention mechanism, a pixel correlationmechanism and multi-scale information. According to the technical scheme provided by the embodiment of the invention, the attention mechanism is effectively utilized to learn the dependency relationship in spatial position and channel dimension, the feature expression capability is enhanced, the pixel correlation mechanism is utilized to make the segmentation result more accurate, and meanwhile,the multi-scale feature information is also utilized to learn a global scene to improve the accuracy of pixel point classification.

The invention discloses a Yellow River ice semantic segmentation method based on a multi-attention mechanism double-flow fusion network. The method is used for solving the technical problem that an existing Yellow River ice detection method is poor in accuracy. According to the technical scheme, firstly, data sets are collected and labeled, the labeled data sets are divided into a training data set and a test data set, then a segmentation network structure is constructed, the network comprises shallow branches and deep branches, and a channel attention module is added to the deep branch, a position attention module is added to the shallow branch, the fusion module is used for fusing the shallow branches and the deep branches, the data in the training set is added into the network in batches, the constructed neural network is trained by adopting cross entropy loss and an RMSprop optimizer, and finally, a to-be-tested image is input and a test is carried out by using the trained model. According to the method, multi-level and multi-scale feature fusion can be selectively carried out, context information is captured based on an attention mechanism, a feature map with higher resolutionis obtained, and a better segmentation effect is obtained.

The disclosure relates to a foreground and background segmentation method and apparatus of an image, and a terminal. The method comprises: a foreground region, a possible foreground region, and a background region of a to-be-segmented image are obtained; a neighbor pixel of each pixel is determined according to color information and spatial position information of each pixel in the to-be-segmented image; on the basis of the foreground region, the possible foreground region, and the background region as well as the neighbor pixel of each pixel, a probability value of each pixel as a target foreground region is determined; and foreground and background segmentation is carried out on the to-be-segmented image based on the probability value of each pixel as the target foreground region in the to-be-segmented image. According to the embodiment of the invention, the neighbor pixel of each pixel can be detected and similarity of all pixels in a full image range can be considered, so that the refined foreground and background segmentation can be realized precisely and the segmentation precision is high.