Information Distance based Photoshop Metric

Objective predictor of perceived modifications to images, or the so-called Photoshop metric, finds many applications and is particularly desirable in the fashion industry. A promising solution to the problem is information distance, which measures the minimal number of bits required to transform one object to another. The application of information distance measures to photo editing assessment, however, is perplexed by human visual system characteristics, such as the difference in attention levels to different parts of an image, and the varying sensitivity at detecting different types of image distortions. Here we make one of the first attempts to develop an information distance based measure for Photoshop metric, where information distance-based features are used to train a Support Vector Regressor (SVR). Experimental results show that the proposed metric is well correlated with mean human observer ratings.