Headdress Detection Based on Saliency Map for Thangka Portrait Image

This paper applies the saliency map of attention model on Thangka image detection for the first time. Making contrast with some detection methods, an automatic detection algorithm is proposed to segment the headdress area of buddha in Thangka portrait image. Firstly, attention model is applied to calculate salient value of each pixel. Then, through using the detection method for back-light area of buddha, a new algorithm which fits for Thangka headdress is proposed. Meanwhile, this research also gives new thought to automatic image segmentation, not only for Thangkaka image, but also for other kinds of research objects. And, experiments show that the detection result is effectively approached to our purpose.