Research On Saliency Detection Based On Multiple Features Fusion And Low Rank Representation

Salient object detection is an important branch in field of computer vision,the purpose of which is to capture the most compelling object area in an image.As a preprocessing method,salient detection is widely used in image classification,image segmentation,object recognition,image compression,image retrieval and image recover and so on.While great progress has been made in salient detection now,there are still a large number of insignificances.For example,in the circumstance of complex background or multi-goals exists in image or salient edge of the object is bright but the interior is not bright enough,the border of the goal can not be well retained and detection is not satisfactory as well.Based on the research of salient detection,the current situation is analyzed and summarized and two kinds of salient object detection methods are given.For the first method,based on visual features,a salient object detection method based on multi-feature fusion is proposed.The definition of three salient features in the method is given:features of continuous edge density,convex hull feature and brightness contrast feature.Multi-features are extracted in sliding detection window generated randomly and fused by naive Bayesian framework model and a significant detection model is established.For the other method,based on the matrix low-rank sparse decomposition theory,a low-rank salient object detection method based on sparse constraints of sub-graphs is proposed.In this method,sub-graphs are obtained by spectral clustering based on super pixels and sparse constraints are imposed on the sub-graphs to emphasize the image spatial structure characteristics.And the higher-rank prior knowledge is added to improve the detection effect.Finally,low-rank sparse decomposition is performed for the model by alternating direction multiplier method to obtain a sparse matrix.Two methods are presented for detecting salient object,the experimental results of which show that the detection has great robustness.Based on the multi-feature fusion detection method,the specific location of the salient object is marked by a rectangular window.As for another method,the sparse constraints on the sub-graph make the salient object separated from the background more completely and the internal salient values are more consistent,which also works well for the image with complex backgrounds.