36results about How to "Full details" patented technology

The invention provides a combined optimizing and training method for the multi-dictionary learning of image sparse representation, which belongs to the multi-media communication and image data processing field, and is characterized by: regarding the non-zero element in the singular value matrix after the singular value decomposition of the gradient matrix of the training image unit as the energy value of the gradient direction; according to set the energy value parameter threshold, dividing the image unit into various isotropic images and various anisotropic images; learning successively a shared dictionary and a specialized dictionary; optimizing the factors to minimize the objective function such as the residuals of the isotropic and anisotropic images after the sparse representation, the autocorrelation and the degree of cross-correlation of each dictionary, and the regularization of nonzero element; in the optimization process, using the orthogonal matching and tracking algorithm to optimize A0 and Ak, and then optimizing D0 and Dk by the gradient descent algorithm; and in keeping the parameters to be optimized, regarding the items not involving the parameters as constants. When the method is used for image compression, the details could be well preserved, a low distortion rate is achieved and a good image quality could also be achieved.

The invention discloses a face figure synthetic method based on coupling neighbor indexes, mainly solving the problem that a present method does not consider the training figure effect in neighbor selection. The method comprises the steps of: 1, diving a training figure sample set, a training photograph sample set and a test sample set; 2, diving all images into image blocks and forming block sets; 3, diving training photograph block sets and corresponding image blocks sets into a plurality of subsets; 4, establishing the indexes of each pair of photograph-image block in the training photograph block subsets and training image block subsets to K neighbor pairs thereof; 5, searching for a nearest neighbor of each test photograph block, and furthermore obtaining K neighbors thereof; 6, utilizing a Markov weight network to solve a reconstruction weight; 7, solving an image block to be synthesized; and 8, iteratively executing steps 5-7 for N times, and obtaining a finally synthesized image through fusion. Compared with the prior art, the synthesized image has higher definition and more complete details, and can be used for face retrieval and identification.

The invention discloses a real-scene three-dimensional model and three-dimensional simulation model hybrid loading method and relates to the field of graphic images. The method is characterized by, to begin with, optimizing a three-dimensional simulation model by utilizing a real-scene three-dimensional model; then, carrying out block division on the real-scene three-dimensional model and the three-dimensional simulation model; thirdly, designing hybrid loading parameters; and finally, carrying out real-scene three-dimensional model and three-dimensional simulation model hybrid loading when camera parameters change in a three-dimensional scene according to the hybrid loading parameters. The method comprehensively utilizes different advantages of two data resources of the real-scene three-dimensional model and the three-dimensional simulation model; and at the macro level, advantages of real color and rich details of an oblique photography model are utilized, and at the micro level, the advantages of complete details and clear textures of three-dimensional simulation are utilized, so that a purpose of carrying out real presentation of a city is achieved, the problem that it is difficult to express city macro and micro forms and colors at the same time by merely utilizing the real-scene three-dimensional model or the three-dimensional simulation model separately is solved, and an application effect of a three-dimensional digital city is improved.

The invention provides a steel plate surface defect image extraction method. The method is characterized by including the following steps that steel plate surface images are subjected to ROI detection, and non-defect images are removed; defect images are sequentially read, the images are preprocessed, the influences of noise and non-uniform illumination on the defect segmentation effect are removed, and the foreground and background contrast ratio is increased; the preprocessed steel plate defect images are subjected to defect region segmentation extraction through Center-Surround Difference, and finally data is output. The efficiency of an image extraction algorithm is improved, it is not needed to occupy large memory space, real-time processing is convenient, the influence of the noise is effectively weakened, wrong extraction, caused by illumination and brightness unevenness, of the images is avoided, the contrast ratio is increased, the defect extraction effect is more excellent, and details are complete.

The invention discloses a method for extracting the boundary of a missing area based on airborne LiDAR point cloud data, comprising the following steps: gridding the original airborne LiDAR point cloud data, obtaining a gridded matrix with missing areas, Define the matrix as a grid; use the seed method to traverse the grid to obtain the missing regions that can be distinguished from each other; expand each missing region to obtain the mesh set surrounding the missing region; use the Kd-Tree method to establish a machine Load the topological relationship between LiDAR point cloud data to obtain the k-neighborhood points to be judged; judge whether the point is a boundary feature point through the distribution uniformity of the point-to-be-judged point and k-neighborhood points; use the nearest point search method to search for scattered boundaries The feature points are connected to obtain the ordered boundary feature points of the missing areas in the airborne LiDAR point cloud data. The invention can obtain the airborne LiDAR point cloud data missing area boundaries with complete details and independent of each other.

The invention discloses a face image synthesis method based on adaptive expression. By using the method, problems that synthesized image definition is low and details are not complete in an existing method are solved. The method comprises the following steps of firstly, processing a database, carrying out image filtering on all the images, then partitioning the images and extracting an image block characteristic so as to obtain one training image block dictionary and two photograph block dictionaries; secondly, testing whether a photograph block includes edge information or face characteristic information, according to that, selecting different dictionaries and searching a neighbor block; and finally, using a Markov network model to obtain image blocks to be synthesized, fusing all the image blocks to be synthesized and acquiring the synthesized image. Compared to the prior art, by using the method, a synthesis result possesses high definition and complete details and can be used for face retrieval and identification.

The invention brings forward a picture renewing method based on resolution improvement, and is applied to pictures with quite low resolution and quite low sharpness. First of all, pictures are preprocessed and trained, secondly, a mapping relation model based on a hybrid expert model is obtained, next, low-resolution pictures to be processed are segmented to blocks, and finally, according to the mapping relation model, the low-resolution pictures are improved to high-resolution pictures. The method provided by the invention has the advantages of good adaptability, fast speed, good effect, high expansibility and the like.

The invention provides a method for reconstructing and displaying a billion-pixel high dynamic range image. The method comprises the step of acquiring a plurality of million-pixel level sub-images, the step of splicing the acquired sub-images into a billion-pixel spherical surface image and adjusting the dynamic range of the spherical surface image, and the step of displaying the spliced image to a user in an interactive mode according to the demands of the user. According to the method, a panoramic image with integral details and different dynamic ranges can be acquired, and the user can integrally take out and display the concerned partial image conveniently.