Saliency Object Extraction In Nature Scene

With the continuous development and application of information technology, image processing has been attracted moreand more researcher. One key problem is at how to find the information we interested in an image data. Human been canroutinely and effortlessly judge the importance of image regions and focus attention on important parts. Therefore, visualsaliency being closely related to how we perceive and process visual stimuli is investigated by multiple disciplines.Detection multiple salient objects is difficult. There are two problems to be solved: on the one hand we try to extract thetarget has a complete semantic structure, and on the other hand, a reasonable mechanism to branch target and determinethe number of interesting targets is needed. To solve this problem, we proposed a new method combining space-basedvisual attention model and object-based visual attention model. We try to use the superiority of these two models to resolvesome common problems in the saliency object extraction. The main contents include:Research on the extraction of saliency object in color natural image which contains multiple targets. Color images containabundant information. First of all, we use the space-based model to get saliency map, then segment the space saliency mapand get different regions. The regions are used to extract salient objects by object-based model. The procedure includesdeciding whether a single region is a single saliency target and judge multiple areas whether it belongs to the samesaliency target, to ensure the extracted target have more complete semantic structure.Research on the extraction of saliency object in gray image. For grayscale image, which contains less information, weproposed an area-based method, extracting multiple targets in the image adaptively: First, use the graph-based approach tosegment the input image into different regions, then calculate the similarity between regions and weighted the similaritywith the distance of regional center, building the saliency map, finally threshold the saliency map to get the saliencyobjects of gray image.Research on target transformation mechanisms of multi-target scenes. Extract the saliency object of image, meanwhile,sort the extracted objects based on their own saliency, then branch the targets according to the ranking results. In addition,we also consider the merge of different region and object, whether it is a color image or grayscale image, the extractedsignificant target after the merge and transfer can get a more accurate objects.Finally, we use the the Achanta image dataset and images in Caltech dataset respectively on the proposed algorithm. Theexperimental results indicate that, the proposed method on multiple saliency objects extraction in static images obtains ansatisfactory result, and the universal applicability of this method make it can be used in many other applications whereredundant information is needed to be filtered out such as information fusion.