60results about How to "Improve segmentation quality" patented technology

The invention discloses an end-to-end difference network learning method for image semantic segmentation. The method comprises the steps that a main network structure and a complete network structureare constructed respectively by using a Caffe deep learning framework, wherein the main network structure is used to generate a rough segmentation model and a small target area of each image in a training set and the complete network is used for final image semantic segmentation; the rough model of the main network is trained by using a part of the data of the training set, and a segmentation result acquired by the rough model is compared with a real segmentation image to acquire a mistaken segmentation area of the rough model; the acquired rough model is used as an initialization parameter totrain a complete network model to acquire a final segmentation result, and an image semantic segmentation model is established; the segmentation model is tested; and all test images are segmented according to the image semantic segmentation model acquired in the step 3. The method can be sensitive to small target areas, and can solve the problems of edge blur and misjudgment of similar parts to some extent.

The present invention relates to a multi-view video image segmentation method based on graph cut and view correlation, which belongs to the field of image processing. With the object to overcome the shortcomings seen in a prior method such as poor robustness, slow execution rate and unsatisfactory segmentation effect, a multi-view video image segmentation method is provided which comprises mainly two parts. An improved Grabcut algorithm is used to segment any single viewpoint image that is given. The introduced smooth item makes it easier to extract a target from the background and the measure to protect edge of the target ensures that the edge of an obtained target is still a smooth one. On the basis of the obtained target of viewpoint image, other targets of viewpoint images are extracted by view disparity matching and a morphological processing method. The algorithm adopts a segmentation method featuring stereoscopic targets matching and tracking in a super-pixel domain, improves the Slic super-pixel generation algorithm and redefines the energy function, which effectively raises the segmentation efficiency under the premise of ensuring a segmentation result with high accuracy and promises the algorithm for wide applications.

The invention provides a superpixel clustering method and equipment based on multiple features, and the method fuses the boundary features and texture features of an image, and achieves the superpixelsegmentation bycombining with the multiple features of the image. The method comprises a clustering stage and a merging stage. The method comprises the steps of:in the clustering stage, judging the affiliation of the pixels by measuring the similarity between the pixels, and clustering to obtain pre-segmented super-pixels; and enabling the similarity measurement of the clustering stage tostart from color features and spatial position features of pixels, and adding boundary feature factors to adjust the similarity degree between pixel points located in a boundary adjacent region. The edges ofthe superpixels obtained in the process are tightly attached, but segmentation is too fine, and therefore the superpixels need to be further corrected. In the merging stage, the similarity degree between the superpixels is measured according to the uniqueness of the content of the superpixels, and the scattered superpixels are aggregated to obtain the final superpixels. Compared with an existing method, the method has the advantage that higher performance is shown in the aspect of keeping the super-pixel boundary and the object boundary in the image fit.

The invention discloses an SMT (Surface Mounting Technology) welding spot image segmentation method, which comprises the steps of 1) conducting preprocessing on an original welding spot RGB (Red-Green-Blue) true color image, including smoothing a welding spot color image, converting a color space from RGB to HIS (Hue-Intensity-Saturation) and sharpening a welding spot image; 2) segmenting the component H (Hue) of the welding spot image, including dividing the welding spot image into n sub-images, solving the segmentation threshold lambda k (k is more than or equal to 1 and is less than or equal to n) of each sub-image by adopting an improved maximum between-cluster variance method and respectively segmenting each sub-image according to the segmentation threshold lambda k of each sub-image; 3) segmenting the component I (Intensity) of the welding spot image by adopting a region growing method; 4) conducting image arithmetic operation on the component H and the component I of the welding spot image to obtain an intact welding spot morphological image; and 5) conducting morphological processing to the intact welding spot image to obtain a final segmented image. By using the SMT welding spot image segmentation, the phenomenon that the traditional segmentation method causes mistaken segmentation and cannot conduct segmentation method can be effectively avoided and the segmentation quality of the welding spot image is improved.

The invention belongs to the technical field of cloth processing and particularly relates to integrated non-woven fabric producing, cutting and winding processing equipment which comprises a horizontal base plate; a fixing mechanism, a compressing mechanism and a driving mechanism are sequentially mounted on the horizontal base plate from right to left; two electric telescopic rods are verticallyand fixedly mounted on the horizontal base plate; a guide rod is horizontally and fixedly mounted between the two electric telescopic rods; and a cutting mechanism is mounted on the guide rod. When the integrated non-woven fabric producing, cutting and winding processing equipment is adopted to cut and wind non-woven fabric, offcut generated at a cutting opening can be sucked and removed, so thatthe offcut is prevented from being adhered to the surface of non-woven fabric, the surface quality of the non-woven fabric is improved, a cutting line can be ensured to be parallel and level, and longitudinal folds on the non-woven fabric can be leveled off to improve the surface evenness.

Embodiments of this application disclose a method for training an image semantic segmentation model performed at a server, to locate all object regions in a raw image, thereby improving the segmentation quality of image semantic segmentation. The method includes: obtaining a raw image used for model training; performing a full-image classification annotation on the raw image at different dilation magnifications by applying a multi-magnification dilated convolutional neural network model to the raw image, and obtaining global object location maps in the raw image at different degrees of dispersion corresponding to the different dilation magnifications, wherein a degree of dispersion is used for indicating a distribution of a target object on an object region positioned by the multi-magnification dilated convolutional neural network model at a dilation magnification corresponding to the degree of dispersion; and training an image semantic segmentation network model using the global object location maps as supervision information.

The invention discloses a video segmentation method based on a statistical shape prior. The method comprises the following steps of: 1) video segmentation initialization; 2) foreground shape matching and alignment, and calculating a statistical shape prior measurement; 3) optimizing video segmentation based on the statistical shape prior measurement; 4) repeating the step 2) and step 3) more than twice and finishing. The invention provides a novel foreground shape matching and alignment method which can effectively extract a correct foreground local similar shape and a correct foreground overall similar shape in a video. The invention further provides a novel shape priori statistical method which can be applied in any video segmentation method in order to improve the segmentation quality. Finally, the invention provides a video segmentation algorithm based on a combination of the statistical shape prior measurement, an overall color probability statistics measurement and a background subtraction measurement, and the algorithm can realize a robust segmentation of the foreground and background which have similar colors.

The invention relates to a multi-focus image edge detection method based on space relative altitude information, belonging to the field of digital image acquisition and identification. The multi-focus image edge detection method comprises the steps of: obtaining a multi-focus sequence image of a target sample by using an image acquisition system; carrying out image registration preprocessing on the multi-focus sequence image by using a cross-correlation information-based image frequency domain registration method; fusing the registered image by using a wavelet transform-based image fusion algorithm and extracting a space relative altitude information image by using a focus depth measuring method; processing the space relative altitude information image by using an altitude consistency constraint method for eliminating noise; carrying out edge detection on a target region and a background region of the processed space relative altitude information image; and identifying and expanding a local maximum region of a sub target in the target region of the detected space relative altitude information image for completing the edge detection of the space relative altitude information image so that the edge detection of the multi-focus image is realized.

The invention discloses an SAR image ground object cutting method based on a random projection and an improved spectral cluster and belongs to the technical field of image processing. The SAR image ground object cutting method based on the random projection and the improved spectral cluster is the tentative research aimed at the application of a compression sensing radar system and is mainly used for directly carrying out a series of processing on observation vectors of a target for image cutting. According to the process, an image to be cut is partitioned through a 9*9 window according to the pixels; one-dimension Gaussian random observation is carried out on each block, and the obtained observation vectors are stored into Y1; the density sensitive distance between the blocks is worked out with each observation vector as a whole, and a Laplacian matrix is constructed; a feature vector corresponding to the maximum feature value of the Laplacian matrix is worked out, and a matrix V is constructed; the row vector of the V is normalized, a matrix X is obtained, each row of the X is regarded as a point, the points are clustered as a cluster k by utilizing an average value K, and each row is marked with a category number; the mark numbers of all pixels in an SAR image are output and displayed in a result image with different colors, and the final cutting result image is obtained. The SAR image ground object cutting method based on the random projection and the improved spectral cluster has a good cutting effect under a low sampling rate, and can be applied to the field of SAR image ground object cutting based on the compression sensing theory.

The invention discloses an image partition method based on neighbor-hood PCA (Principal Component Analysis)-Laplace. The method comprises the steps of carrying out PCT on an original image to obtain a feature vector of each pixel; extracting the main ingredients of an image to effectively restrain noise; carrying out edge detection on the image by a Laplace operator so as to realize image segmentation. Compared with a traditional Sobel operator partitioning algorithm and an LOG operator partitioning algorithm, the method is used for carrying out the PCA on the image pixels to estimate parameter values in a de-noising process in a way of being independent of an empirical value, so as to effectively reduce the interference on the image by the noise, and simplify the computation complexity. Experimental results show that the method can improve the segmentation effect of the image and is strongly excellent in accuracy and robustness.

The invention discloses a fuzzy C-kernel mean clustering segmentation method based on improved whale algorithm optimization, and the method comprises the steps: inputting an image, setting parameters,initializing whale positions, calculating the adaptive value of each whale, determining an asynchronous communication mechanism, and finally outputting the optimal whale position. According to the method, an asynchronous communication strategy and a selection mechanism are introduced to improve the whale algorithm, so that the convergence speed and precision of the algorithm are further improved;and segmenting the synthetic aperture radar image. A test result shows that the algorithm has good segmentation quality and can realize rapid segmentation of the SAR image at the same time.

According to a high-resolution SAR image segmentation method for improving the FCM through multi-stage cooperation, the theoretical basis of an FCM clustering algorithm and the limitation of the FCM clustering algorithm for high-resolution SAR image segmentation are analyzed, weighted median filtering improvement is conducted on the FCM in combination with the spatial neighborhood relation of pixels, aiming at the limitation that a current FCM segmentation result is prone to falling into local optimum, SA is improved to further optimize the FCM, the effectiveness of the improved and optimized algorithm is verified through experiments, finally, the segmentation result of the improved and optimized FCM clustering algorithm is used as the initial segmentation of maximum posterior probability superposition re-segmentation, the maximum posterior probability superposition SAR re-segmentation is adopted to segment the image, qualitative and quantitative comparative analysis is carried out on the segmentation results of the above methods to obtain a series of improvements, the SAR segmentation quality is obviously improved, details and contours of image edge areas are clear, segmentation is accurate, robustness and reliability are good, meanwhile, the resistance of the algorithm is enhanced, and the SAR segmentation quality and efficiency are greatly improved.

The invention provides an SAR image registration method based on an MRF image segmentation algorithm, and is used for solving the technical problem that an existing feature-based SAR image registration method is low in registration efficiency and poor in stability. The SAR image registration method comprises the following steps: segmenting an SAR image reference diagram and a to-be-registered diagram by using the MRF image segmentation algorithm; carrying out regional interception on the segmented reference diagram and the segmented to-be-registered diagram; respectively corresponding the intercepted segmented reference diagram image blocks and the intercepted segmented to-be-registered diagram image blocks into an SAR image reference diagram and a to-be-registered diagram; establishing SAR-Harris scale space of the SAR image reference diagram image blocks and SAR-Harris scale space of the to-be registered diagram image blocks; extracting SARSIFT feature points of the reference diagram image blocks and the to-be-registered diagram image blocks to form a matching point pair set; removing mismatched points by using a RANSAC algorithm; and optimizing the matching point pair set by using a mutual information method.

The invention discloses a cancer cell HER2 gene amplification analyzing method and system. According to the invention, the method comprises: carrying out red dot cluster analysis; carrying out red dotrecognition; carrying out green dot recognition; constructing a cell nucleus character matrix; carrying out segmentation to obtain an area and a circularity value of a cell nucleus; calculating a mean ratio value, a mean green dot value, and a mean red dot value of each cell; determining the HER2 gene is amplified; and outputting a FISH test result report based on determination of HER2 gene amplification. According to the invention, the HER2 gene amplification detection and analysis are carried out by using the cancer cell FISH image; with simulation of the pathologist diagnostic process, thered-dot and green-dot ratio values of the FISH image can be identified effectively by the computer and thus whether the HER2 gene is amplified is determined; and thus the workload for the pathologistis reduced.

The invention discloses a fast segmentation method for a grayscale image histogram based on K-harmonic means clustering. The fast segmentation method comprises the steps of inputting a grayscale image, and calculating to acquire a grayscale histogram; determining the clustering segmentation number m of the image, and setting the value of a K-harmonic means power exponent p; calculating a new membership function value matrix U(t+1) according to a KHM (K-harmonic means) clustering principle; calculating an image weighting function vector W; calculating a newly acquired clustering center vector Cj; stopping algorithm iteration if the norm U(t+1)-U(t) is less than epsilon or reaches an allowable maximum iteration value; if not, supposing t=t+1, and carrying out algorithm iteration continuously; and segmenting the image according to a membership matrix U(t) acquired by convergence. The fast segmentation method disclosed by the invention is not sensitive to a clustering initial value, the stability is high, and the segmentation quality of the grayscale image is improved. In addition, the fast segmentation method is high in time efficiency, capable of realizing a real-time segmentation task, and applicable to task occasions with a high real-time requirement.

The invention discloses a K-S distance merging cost-based SAR image partitioning method, and mainly aims to solve the problems of low partitioning speed and poor partitioning quality on an SAR image in the prior art. The technical scheme comprises the steps of 1, calculating proportional edge strength mapping RESM (x, y) according to the pixel value of an original SAR image, and performing watershed transform to obtain an initial partitioning result; 2, calculating an empirical distribution function Fn(x) of each region pixel value in the initial partitioning result; 3, calculating a K-S distance KSD (a, b) of any two adjacent regions a and b in the initial partitioning result; 4, calculating a value of a merging cost function K(a, b); and 5, determining a final image partitioning result according to the value of the merging cost function K(a, b). By virtue of the method, the SAR image partitioning quality is improved, the complexity is relatively low, the operating speed is relatively high, and the method can be used for SAR image partitioning in a complex scene.

The invention discloses an information processing method and device and an electronic device. The method comprises the steps of obtaining the traffic information of at least one reference intersectionin a target area of a preset area and the traffic information of a target intersection having a set area relationship with the target area; according to the traffic information of the reference intersection and the traffic information of the target intersection, determining a delay index of a new area formed by the target area and the target intersection as a first delay index under the conditionthat the target intersection is not divided into the target area; according to the traffic information of the reference intersection and the traffic information of the target intersection, determining a delay index of a new area as a second delay index under the condition that the target intersection is divided to the target area; and determining whether to divide the target intersection into thetarget area according to the first delay index and the second delay index.

The invention discloses a gray scale image threshold segmentation method based on symmetric Gamma divergence, and the method comprises the steps: inputting a to-be-segmented image, and solving a normalized gray scale histogram of the to-be-segmented image; building a symmetric Gamma divergence expression of the image before and after segmentation; solving a gray scale value which enables the gray scale value of the expression to be minimum in a gray scale range of the image; carrying out the threshold segmentation of the image through the gray scale value, and outputting a segmented image. The method improves the quality of image segmentation. The edge contour of the segmented image is precise, and the texture detail is clear. The method improves the universality, and is suitable for an image processing task with high requirements for instantaneity.

The invention brings forward a graphic text image segmentation method and system based on a line cutting direction. Since a target in a graphic text image has a thin and long structure, the target exhibits the noticeable directivity feature, that is to say, the target exhibits relatively high correlation in a line cutting direction. The local area of the target is increased in the cutting direction of lines by utilizing the feature of the thin and long structure of the target, analyzing the local features of the target in the line cutting direction and also combining a dynamic area increase process during segment of a graphic text image, so line fracture in a segmentation result is effectively minimized and the segmentation quality of the image is improved.

The invention discloses a rice grain segmentation method, a terminal and a storage medium, and the method comprises the steps: carrying out the noise reduction and binarization of a collected rice grain image, carrying out the connected region detection of an obtained binarized image after noise reduction, and obtaining a connected region; according to a preset threshold value and the perimeter and the area corresponding to the communication area, dividing the communication area into an adhesion rice area or a single-grain rice area; obtaining a candidate clustering center pixel point in the adhesion meter area according to the coordinate value of each pixel point in the adhesion meter area and a density peak clustering algorithm; training a preset Gaussian mixture model according to an EMalgorithm, a contour coefficient algorithm, the coordinate value of each candidate clustering center pixel point in the adhesion meter area and the coordinate values of other pixel points to obtain an optimized Gaussian mixture model; and carrying out rice grain segmentation on the adhesion rice area according to the optimized Gaussian mixture model to obtain a segmentation result of the adhesionrice area. According to the invention, the problem of poor segmentation quality of the rice grain image in the existing adhesion object segmentation method is solved.