Beijing University of Technology

Lijuan Duan

Wei Ma

Jun Miao

Jia Qi Li

Shuo Liu

Bin Yuan

Fangfang Liang

Song Cui

Yuanhua Qiao

Shuo Qiu

Junfa Liu

The present invention relates to a clustering method based on iterations of neural networks, which comprises the following steps: step 1, initializing parameters of an extreme learning machine; step 2, randomly choosing samples of which number is equal to the number of clusters, each sample representing one cluster, forming an initial exemplar set and training the extreme learning machine; step 3, using current extreme learning machine to cluster samples, which generates a clustering result; step 4, choosing multiple samples from each cluster as exemplars for the cluster according to a rule; step 5, retraining the extreme learning machine by using the exemplars for each cluster obtained from step 4; and step 6, going back to step 3 to do iteration, otherwise obtaining and outputting clustering result until clustering result is steady or a maximal limit of the number of iterations is reached. The present invention resolves problems that how to realize clustering of high dimensional and nonlinear data space and that the prior art consumes a larger memory or need longer running time.

Clustering method based on iterations of neural networks

A method of establishing a 3D saliency model based on 3D contrast and depth weight, includes dividing left view of 3D image pair into multiple regions by super-pixel segmentation method, synthesizing a set of features with color and disparity information to describe each region, and using color compactness as weight of disparity in region feature component, calculating feature contrast of a region to surrounding regions; obtaining background prior on depth of disparity map, and improving depth saliency through combining the background prior and the color compactness; taking Gaussian distance between the depth saliency and regions as weight of feature contrast, obtaining initial 3D saliency by adding the weight of the feature contrast; enhancing the initial 3D saliency by 2D saliency and central bias weight.

Establishment method of 3D saliency model based on prior knowledge and depth weight

The present invention discloses a method for detecting a salient region of a stereoscopic image, comprising: step 1) calculating flow information of each pixel separately with respect to a left-eye view and a right-eye view of the stereoscopic image; step 2) matching the flow information, to obtain a parallax map; step 3) selecting one of the left-eye view and the right-eye view, dividing it into T non-overlapping square image blocks; step 4) calculating a parallax effect value for each of the image blocks of the parallax map; step 5) for each of the image blocks of the selected one of the left-eye view and the right-eye view, calculating a central bias feature value and a spatial dissimilarity value, and multiplying the three values, to obtain a saliency value of the image block; and step 6) obtaining a saliency gray scale map of the stereoscopic image from saliency values of the image blocks. The present invention provides a method for extracting stereoscopic saliency based on parallax effects and spatial dissimilarity, acquiring depth information by utilizing parallax, and combining visual central bias feature and spatial dissimilarity to realize more accurate detection of a stereoscopic salient region.

Method for detecting salient region of stereoscopic image

The present invention relates to a method for detecting visual saliencies of a video image based on spatial and temporal features, including: dividing an input image into image blocks and vectorizing the image blocks; decreasing dimensions of each image block through principal component analysis; calculating a dissimilarity between each image block and each of the other image blocks; calculating a visual saliency of each image block by combining a distance between image blocks, to obtain a spatial feature saliency map; imposing a central bias on the spatial feature saliency map; calculating a motion vector of each image block, extracting a temporal visual saliency of the current image by combining motion vectors of previous two frames, to obtain a temporal feature saliency map; integrating the spatial feature saliency map and the temporal feature saliency map to obtain a spatiotemporal feature saliency map, and smoothing the spatiotemporal feature saliency map to obtain a resulted image finally reflecting a saliency of each region on the current image. In the present invention, a saliency map integrating the temporal features and the spatial features, so that saliencies in different regions in a video may be predicted more accurately.

Method for detecting visual saliencies of video image based on spatial and temporal features

A method for embedding and extracting a multi-scale space based watermark, comprises: constructing a pyramid structure of an original image by dividing each carrier image layer into M square carrier image blocks of the same size; constructing a multi-scale structure of a watermark image; embedding a watermark by embedding each watermark image into a corresponding carrier image block to obtain the original image containing the watermark; locating in the pyramid structure of the original image a target image from which a watermark will be extracted; extracting the watermark by obtaining an estimated watermark by means of the target image block and the reference image block; comparing watermarks by evaluating similarity between the estimated watermark and a watermark image to which the reference image block corresponds. Due to the multi-resolution block pyramid data structure in the present invention, a large scale attack is decomposed into a multi-level small scale attack.

Growing community of inventors

Beijing, China

Lijuan Duan