Abstract: An image feature extraction method for person re-identification includes performing person re-identification by means of aligned local descriptor extraction and graded global feature extraction; performing the aligned local descriptor extraction by processing an original image by affine transformation and performing a summation pooling operation on image block features of same regions to obtain an aligned local descriptor; reserving spatial information between inner blocks of the image for the aligned local descriptor; and performing the graded global feature extraction by grading a positioned pedestrian region block and solving a corresponding feature mean value to obtain a global feature. The method can resolve the problem of feature misalignment caused by posture changes of pedestrian, etc., and eliminate the effect of unrelated backgrounds on re-recognition, thus improving the precision and robustness of person re-identification.

Abstract: Disclosed is a deep discriminative network for person re-identification in an image or a video. Concatenation are carried out on different input images on a color channel by constructing a deep discriminative network, and an obtained splicing result is defined as an original difference space of different images. The original difference space is sent into a convolutional network. The network outputs the similarity between two input images by learning difference information in the original difference space, thereby realizing person re-identification. The features of an individual image are not learnt, and concatenation are carried out on input images on a color channel at the beginning, and difference information is learnt on an original space of the images by using a designed network. By introducing an Inception module and embedding the same into a model, the learning ability of a network can be improved, and a better differentiation effect can be achieved.

Abstract: A bidirectional image-text retrieval method based on a multi-view joint embedding space includes: performing retrieval with reference to a semantic association relationship at a global level and a local level, obtaining the semantic association relationship at the global level and the local level in a frame-sentence view and a region-phrase view, and obtaining semantic association information in a global level subspace of frame and sentence in the frame-sentence view, obtaining semantic association information in a local level subspace of region and phrase in the region-phrase view, processing data by a dual-branch neural network in the two views to obtain an isomorphic feature and embedding the same in a common space, and using a constraint condition to reserve an original semantic relationship of the data during training, and merging the two semantic association relationships using multi-view merging and sorting to obtain a more accurate semantic similarity between data.

Abstract: The present application provides a filtering method for intra-frame and inter-frame prediction. For a pixel value obtained by prediction, whether to filter a prediction block is self-adaptively determined by means of an RDO decision; and a filter is established based on a generalized Gaussian distribution function, and different filters are used for prediction blocks of different sizes, and filter parameters are obtained by training the prediction blocks of each size. The method comprises performing RDO on each coding unit on a coding end, and firstly performing intra-frame prediction or inter-frame prediction on each coding unit; then filtering the prediction block using the filtering method of the present invention, and determining whether to use prediction block filtering by means of an RDO decision; on a decoding end, configuring different filtering coefficients for different prediction block sizes; and decoding according to a flag.

Abstract: The present application provides a filtering method for intra-frame and inter-frame prediction. For a pixel value obtained by prediction, whether to filter a prediction block is self-adaptively determined by means of an RDO decision; and a filter is established based on a generalized Gaussian distribution function, and different filters are used for prediction blocks of different sizes, and filter parameters are obtained by training the prediction blocks of each size. The method comprises performing RDO on each coding unit on a coding end, and firstly performing intra-frame prediction or inter-frame prediction on each coding unit; then filtering the prediction block using the filtering method of the present invention, and determining whether to use prediction block filtering by means of an RDO decision; on a decoding end, configuring different filtering coefficients for different prediction block sizes; and decoding according to a flag.

Abstract: An intra code-rate predicting method in video coding, which enables skipping of an entropy coding procedure during a rate-distribution optimization process and effectively reduces the coding time by modeling residual information of a prediction block and predicting the number of coding bits based on a model resulting from the modeling.

Abstract: Disclosed is a virtual viewpoint synthesis method based on image local segmentation, which relates to the digital image processing technology.

Abstract: A Collaborative Deep Network model method for pedestrian detection includes constructing a new collaborative multi-model learning framework to complete a classification process during pedestrian detection; and using an artificial neuron network to integrate judgment results of sub-classifiers in a collaborative model, and training the network by means of the method for machine learning, so that information fed back by sub-classifiers can be more effectively synthesized. A re-sampling method based on a K-means clustering algorithm can enhance the classification effect of each classifier in the collaborative model, and thus improves the overall classification effect.

Abstract: The present disclosure provides a hybrid framework-based image bit-depth expansion method and device. The invention fuses a traditional de-banding algorithm and a depth network-based learning algorithm, and can remove unnatural effects in an image flat area whilst more realistically restoring numerical information of missing bits. The method comprises the extraction of image flat areas, local adaptive pixel value adjustment-based flat area bit-depth expansion and convolutional neural network-based non-flat area bit-depth expansion. The present invention uses a learning-based method to train an effective depth network to solve the problem of realistically restoring missing bits, whilst using a simple and robust local adaptive pixel value adjustment method in an flat area to effectively inhibit unnatural effects in the flat area such as banding, a ringing and flat noise, improving subjective visual quality of the flat area.

Abstract: The disclosure is directed to methods, apparatuses, and devices for synthesizing virtual viewpoint images that are configured for acquiring an original image set, performing stereo matching to original images to obtain a first disparity map set, performing optimization processing to each first disparity map in the first disparity map set to obtain a second disparity map set, determining whether each second disparity map in the second disparity map set has a third disparity map with deep tomography to obtain a first determination result, if the first determination result indicates existence of the third disparity map, segmenting a portion with deep tomography to obtain a corresponding foreground disparity map and a corresponding background disparity map, mapping the second disparity map without deep tomography, the foreground disparity map, the background disparity map, and the portion without deep tomography in the third disparity map according to a first virtual viewpoint to obtain a first virtual viewpoint

Abstract: Disclosed is a panoramic image mapping method, wherein mapping regions and a non-mapping region are partitioned for an equirectangular panoramic image with a resolution of 2M×M, where only the partitioned mapping regions are mapped as square regions; the method comprises:, computing a vertical distance and a horizontal distance from a point on the square region to a center of the square region, a larger one of which being denoted as m; computing a distance n from the point to a zeroth (0th) point on a concentric square region; computing a longitude and a latitude corresponding to the point; computing a corresponding position (X, Y) in the equirectangular panoramic image to which the point is mapped; and then assigning a value to the point. The method may effectively reduce oversampling, thereby effectively reducing the number of pixels of the panoramic image and the code rate required for coding with little distortion.

Abstract: Disclosed is a panoramic image mapping method and a corresponding reversely mapping method. Particularly, the mapping process includes mapping a panoramic image or a spherical surface corresponding to Video A: first, dividing the spherical surface into three areas based on the latitudes of the spherical surface, denoted as Area I, Area II, and Area III, respectively; mapping the three areas to a square plane I?, a rectangular plane II?, and a square plane III?, respectively; then, splicing the planes I?, II? and III? into a plane, wherein the resulting plane is the two-dimensional image or video B. Compared with the equirectangular mapping method, the method according to the present disclosure may effectively ameliorate oversampling in high-latitude areas and effectively lower the bit rate needed by coding and the complexity of decoding. The present disclosure relates to the field of virtual reality, which may be applied to panoramic images and videos.

Abstract: Disclosed is a boundary filtering method for intra prediction, relating to the video encoding technology filed. Whether boundary filtering is performed on an intra prediction block or not is adaptively selected by means of a rate distortion optimization decision; during filtering, a filter coefficient exponentially attenuated relative to distance to boundary is adopted to perform filtering on the first N rows or the first N columns of the intra prediction block by means of an intra prediction block filter, and different filtering strengths are used according to different sizes of the prediction blocks. Therefore, the boundary distortion problem of intra prediction block is solved, the intra prediction precision is improved, and the encoding efficiency of intra prediction block is increased; and the practicability and the robustness of the boundary filtering technology are improved.

Abstract: An intra-frame and inter-frame combined prediction method for P frames or B frames. The method comprises: self-adaptively selecting by means of a rate-distortion optimization (RDO) decision whether to use the intra-frame and inter-frame combined prediction or not; using a method for weighting an intra prediction block and an inter prediction block in the intra-frame and inter-frame combined prediction to obtain a final prediction block; and obtaining the weighting coefficient of the intra prediction block and the inter prediction block according to prediction distortion statistics of the prediction method. Therefore, prediction precision can be improved, and coding and decoding efficiency of the prediction blocks are improved.

Abstract: An image deblurring method based on light streak information in an image is provided, wherein shape information of a blur kernel is obtained based on a light streak in a motion blur image and image restoration is constrained by combining the shape information, a natural image and the blur kernel to thereby obtain an accurate blur kernel and a high-quality restored image. The method specifically comprises: selecting an optimum image patch including an optimum light streak; extracting shape information of a blur kernel from the optimum image patch including the optimum light streak; performing blur kernel estimation to obtain the final blur kernel; performing non-blind deconvolution and restoring a sharp restored image as a final deblurred image. The present disclosure establishes a blurry image test set of captured images including light streaks and a method to obtain an accurate blur kernel and a high quality restore image.

Abstract: Disclosed is an intraframe prediction method, wherein for an intraframe predicted block in an interframe predicted frame, when an interframe predicted blocks exists on the lower right of the intraframe predicted block, a coding and decoding sequence of intraframe and interframe blocks is adjusted, and intraframe prediction is performed in a lower right intraframe prediction approach, i.e.

Abstract: A cross-media search method using a VGG convolutional neural network (VGG net) to extract image features. The 4096-dimensional feature of a seventh fully-connected layer (fc7) in the VGG net, after processing by a ReLU activation function, serves as image features. A Fisher Vector based on Word2vec is utilized to extract text features. Semantic matching is performed on heterogeneous images and the text features by means of logistic regression. A correlation between the two heterogeneous features, which are images and text, is found by means of semantic matching based on logistic regression, and thus cross-media search is achieved. The feature extraction method can effectively indicate deep semantics of image and text, improve cross-media search accuracy, and thus greatly improve the cross-media search effect.

Abstract: A method for image matching includes acquiring a template image and a target image; acquiring a group of template features according to the template image; extracting a group of target features according to the target image; and according to template features and target features, calculating an degree of image similarity between the template image and each frame of target images, and using a target image with the maximum degree of image similarity as a matched image to the template image. In the image-matching method, image matching is performed by calculating an degree of image similarity between a template image and each target image according to a degree of image similarity between template features and target features, so that non-redundancy of features in an image matching process and correct image matching can be guaranteed, and the image matching accuracy can be improved.

Abstract: Disclosed are a describing method and a coding method of panoramic video ROIs based on multiple layers of spherical circumferences. The describing method comprises: first setting a center of the panoramic video ROIs; then setting the number of layers of ROIs as N; obtaining the size Rn of the current layer ROI based on a radius or angle; obtaining the sizes of all of the N layers of ROIs, and writing information such as the center of the ROIs, the number of layers, and the size of each layer into a sequence header of a code stream. The coding method comprises adjusting or filtering an initial QP based on a QP adjusted value and then coding an image. By flexibly assigning code rates to multiple layers of panoramic video ROIs, while guaranteeing a relatively high image quality of ROIs, the code rate needed for coding and transmission is greatly reduced.

Abstract: Disclosed is a panoramic image mapping method, wherein mapping regions and a non-mapping region are partitioned for an equirectangular panoramic image with a resolution of 2M×M, where only the partitioned mapping regions are mapped as square regions; the method comprises:, computing a vertical distance and a horizontal distance from a point on the square region to a center of the square region, a larger one of which being denoted as m; computing a distance n from the point to a zeroth (0th) point on a concentric square region; computing a longitude and a latitude corresponding to the point; computing a corresponding position (X, Y) in the equirectangular panoramic image to which the point is mapped; and then assigning a value to the point. The method may effectively reduce oversampling, thereby effectively reducing the number of pixels of the panoramic image and the code rate required for coding with little distortion.