The regions of an interest image are the parts of priority attention and have more significance. The traditional visual attention model describes the information of regions of interest using fixed-size circles and can′t accurately express the outline of the regions of interest. A new automatic detection algorithm of the regions of interest based on adaptive radius search (ARS) is proposed. The new algorithm extracts color, intensity and orientation features of the image to generate a multi-scale saliency map. The global saliency threshold is calculated, which can get the end condition of searching the focus of attention. The adaptive radius search mechanism based on saliency ratio is proposed in the description of regions of interest to acquire the accurate information of regions of interest. The experimental results show that the new algorithm not only can effectively improve the detection precision of regions of interest, but also is more suitable to the features of human visual system. It has important value for the automatic target recognition of regions of interest in the future.