Perceptual picture emphasis using texture power maps

Abstract

Applying selective emphasis to photographs is a critical aspect of the visual design process. There is evidence from psychophysics that contrast in texture is a key contributor to saliency in an image, yet unlike other low-level perceptual features, texture cannot be directly modified with existing image-processing software. We present a post-processing technique to subtly change the salience of regions of an image by modifying spatial variation in texture. Our method is inspired by computational models of visual attention that capture sensitivity to outliers in local feature distributions. We use the steerable pyramid, which encodes multi-scale oriented image features and compute a set of power maps which capture the local texture content at each scale and orientation. With this representation, texture variation can be modified to selectively add or remove emphasis in the image. Two user studies provide qualitative and quantitative psychophysical validation of our approach.