A Computational Model Of Visual Attention Based On Feature Combination

With the increasing development of information technology, especially the Inter-net, we are confronted with severe challenges about the large scale data. These data consists not only the simple text data, but also more multimedia data, such as images, videos, and so on. Processing these large data in real-time is an extremely daunting task. How to select the useful and tractable data from the large scale data is an impor-tant research topic with great practical value.Attention is an important cognitive psychology concept originally, which is a s-electing mechanism of the human brain processing the information from the outside world. To solve the compression problem of large scale data, we establish the compu-tational model of visual attention to simulate the human’s selecting mechanisms. The traditional models of visual attention are mostly based on bottom-up features, which just contain the low-level scene information. The result of these models have a bias compared with the human eye, because there is no guidance of the top-down priori knowledge. Guided by these considerations, this paper aims to develop a new compu-tational model of visual attention.The main works of this paper are:(1) Combining the top-down features into the model, such as the area of the atten-tional region and the distance from the attentional region to the center of scene. The effect of this model more corresponds to the observations of the human eye.(2) Proposing a good normalization strategy. The method proposed in this paper solves the problem about combining the bottom-up features with top-down features, and also enriches our understanding of the Feature Integration Theory. (3) Introducing the additional concepts about the mean and variance of regional saliency, makes the computational model have the good robustness. Because it solves a problem about selecting erroneous attentional focus with the large saliency.(4) Modifying the evaluation criteria for attention model. Besides the evaluation of the results of attentional transferring, selecting the first focus of attention is also taken into account.