Research And Application Based On Image Target Saliency Detection

At present, the existing image visual saliency analysis algorithm, which is based on different theoretical methods, but on the effect of its significant target extraction, are focused on the analysis of the overall and the difference of the significant target. And, in the measurement and calculation of the salient region, the overall and local angle reflect different characteristics, the comprehensive consideration of consistency and local differences, can accurately grasp the comprehensive characteristics of the region. If we could obtain the overall and local information from image, we could accurately describe the salient targets of the image.Based on the above analysis, this paper is based on the angle of global consistency and local difference. We agree the view that IG algorithm can keep good target integrity, but ignore the contrast of the local detail, and suppresses the significant region. So, an new NIF algorithm is introduced to highlight the local feature difference. In this paper, respectively detect the saliency information based on global consistency and local differences. Finally, an new method named MAX is formed by combining the above two methods, which can achieve the effect of the whole and local differences. And the significant information of the overall consistency and local differences are detected respectively. Finally, the combination of the two is a new method to detect the significance of MAX algorithm, so as to achieve the effect of the global consistency and local difference. Secondly, this paper proposes a new segmentation method which based on saliency information. To avoid over-segmentation or other segmentation problems, a clustering algorithm is used to region merging and final segmentation results. The final experimental results are satisfactory.Experiments on the images of the nature image database show that the proposed saliency detection algorithm has the same effect in comparison with the artificial segmentation map. When evaluated using one of the largest publicly available dataset, our method outperforms several existing salient object detection methods. And in comparison with other popular algorithms, our algorithm has higher efficiency on application of image segmentation.