Abstract: Saliency object detection means that the computer automatically recognizes the saliency object in the image through the algorithm, which is widely used in many applications such as object recognition, image retrieval and image classification. Aiming at the acquisition of low-rank transformation matrix, the processing of foreground sparse matrix and the relationship between superpixel blocks in the existing significance detection model based on sparse and low-rank matrix restoration, the existing sparse and low-rank matrix restoration model is optimized to make it better applicable to the significance detection of images. Firstly, the low-rank background dictionary is obtained according to the principle of contrast and connectivity. Meanwhile, we used three scales to split multiple feature matrix of images to obtain the foreground sparse matrix of image. Secondly, the structural constraints are made for the results of the significance graph by calculating the influence factor matrix and the confidence matrix between the neighbor pixels, and sparse and low-rank matrix recovery models are used to detect the significance of the image. Finally, the propagation mechanism of K-means clustering algorithm is used to optimize the significant graph. The experimental verification on several datasets shows that the proposed method can accurately and effectively detect saliency object.

Key words: saliency detection, sparse low-rank matrix recover, superpixel