Research On Saliency Detection Algorithm Based On Ant Colony Optimization

Saliency detection is an image and video processing method which was first started in biological area and introduced into computer field in the 1990s.Saliency detection algorithms can be classified into two groups based on human visual system,namely the data driven bottom-up algorithm which is task independent and the knowledge driven top-down algorithm which depends on task.Saliency detection is of great importance on the automatic processing of images and videos.It has been applied into areas including image segmentation,image adaptive compression,image configuration,image non-realistic rendering etc.Image processing can be more precise and efficient with the saliency information.Although there have been excellent results on image saliency detection research,the models on video saliency detection are still limited.With the growing demand on intelligent image and video processing and usage of saliency detection in more fields,more advanced saliency detection technology shows to be promising.This paper reconsiders the relationship between visual saliency detection and the ant colony optimization algorithm which is inspired by the biological phenomenon of ants seeking for food and proposed a novel saliency detection model which is based on ant colony optimization and combines traditional saliency detection conception with it.Experiments are conducted on the proposed model and several other classical saliency detection algorithms to realize a comparison on algorithm performance,advantages and disadvantages.Based on the experimental results,some improvement schemes are proposed and conducted and finally,a novel visual saliency detection algorithm based on ant colony optimization should be proposed.The proposed algorithm is illustrated in two aspects,including research on saliency detection of uncompressed images or videos and on that of compressed videos.As to uncompressed images and videos,we start from feature extraction and take saliency detection experiments on different features and multiple scales.Thus the best model is filtered for experiments on standard database to compare with other classical methods.The evaluation results are given for analysis.Besides,for research on saliency detection of compressed videos,we first extract features including luminance,chrominance,texture and motion vectors directly from the compressed bitstream for single-feature saliency evaluation.Then the saliency detection results of different features are fused together in a way fitting human visual system to obtain the final saliency detection result.In addition,method of multiple scales is also used in this process.In the end,the comparison of proposed method and classical saliency detection methods is conducted for performance analysis.The experimental results exhibits the validity and reliability of the proposed method.