Content-Based cropping using visual saliency and blur detection

Abstract

This research presents an automatic image/frame cropping scheme to preserve the regions of interest in imagery data. First, a blur detection based on Structural Similarity (SSIM) is proposed to identify whether an image contains a blurred background and the sharp foreground objects can then be extracted. The visual saliency is further calculated to help remove insignificant boundaries. Some pre-defined rules are employed to determine more appropriate cropping limits. If further reduction of resolution is necessary, the resulting image after cropping will be scaled directly to the target size. The experimental results show that the proposed method is computationally efficient and the promising results can be achieved in still images and video frames.