GPU Acceleration of Saliency Detection Algorithm

Saliency detection is an important step in image processing on computer, which has been widely used in network graphics, fingerprint recognition and other fields. The method of saliency detection based on structural similarity theory uses the structural similarity to abstract high-level human visual system, measuring the saliency of images through a new Center-Surround operator. It overcomes the mosaic phenomenon of Itti algorithm caused by near interpolation. However, the computational overhead is very large as each pixel needs to be processed. In this paper, we adopt GPU to accelerate this algorithm, which speeds up over 90X compared with original method running on CPU. To the best of our knowledge, the approach we proposed that using GPU to accelerate the saliency detection algorithm is the first one in the field of image processing.