The Algorithm Of Visual Saliency Region Extraction

With the development of information technology, the digital media has become an important media for people to communicate and transfer message. The image resolution is gradually increasing, the image resources is more and more rich, and the image data is facing huge crisis, which all resulting the image information processing, analysis and transmission are required higher and higher accuracy. The unified treatment of the whole image, not only cause high complexity of the algorithm, but also can not guarantee the key area in the image to get the good result. Therefore, we proposes a method for extracting image salient regions in this thesis, which are extracted in multi scales, aims to reduce the complexity of subsequent image processing, and ensure good visual effect.Based on the basic theory of visual attention mechanism and the classic method for extracting significant image, we analysis the image saliency information in the frequency domain and discuss the significant information in the image on the phase spectrum and amplitude spectrum respectively. Reconstruct the image by using the Gauss white noise amplitude spectrum instead of the original in order to simplify the calculation of spectrum residual method. We present an approach to extract salient regions based on frequency domain on multiple-scales, which transform and extract significant information at different resolution scales. Simulation experiments on 58 natural images shows that the proposed method has good visual effect, which can keep the edges of salient region and highlight the whole significant target uniformly at the same time. The area under ROC curve of these results also has satisfied performance compared with other salient region extraction algorithms.Apply the salient region extraction results in image processing applications, such as,image segmentation, extraction and calibrate the initial center of salient region in video tracking algorithm, content aware image resizing and no reference quality assessment. The salient region extraction technique used in image processing, can largely reduce manual participation, improve the visual effect and processing accuracy, also provides a great convenience for the subsequent image processing.