Multiple Visual Features Based Image Complexity Evaluation And Applications

Evaluating the complexity of an image is very important in the fields of esthetics, visual psychology and cognitive science due to that image complexity reflects a fundamental aspect of human vision system. An effective approach to measure it may be a stepping-stone for better understanding of human recognition process and promote the development of other science. Visual complexity is not only related to image content itself, but also determined by how we perceive and process visual stimuli. Therefore, defining an effective complexity measure for color images remains a challenging task. Conventional approaches generally build upon information theory or a certain visual feature. In this paper, we propose a new method by exploring multiple effective visual features including color, clutter and number of objects to measure the complexity of color images.Our basic idea for assessing image complexity is to choose effective features reflecting the complexity of image content and define a quantitative metric to measure the complexity. Following this idea, the proposed solution mainly includes two phases. The first phase is to extract effective features according to psychology of vision. Based on the theory that visual complexity is sensitive to color, clutter and amount of objects, color histogram and gradient orientation histogram is used to measure color complexity and clutter. The intuition behind the cues is that the more complex the image looks, the more histogram-equalized the image should be. In terms of the number of the objects in image, we propose a fuzzy clustering based approach to compute the complexity. The second phase presents a fuzzy clustering model combining the features and defines a complexity class to measure the image complexity. In order to compare two images in the same complexity class, we also calculate a specific score using an internal mapping function.Different from traditional image complexity measures, this paper presents a new comprehensive image complexity measure. For image feature extraction, we extract three key visual features’complexity quantifying the color, clutter and amount of objects. Comparing with the methods based on a single feature, multiple features based image complexity evaluation is proved more robust. Meanwhile, experiments show good consistency between the proposed objective metric and subjective assessment by human observers. In terms of the results, we provides two presentation forms:the image complexity classes denoted by VL, VM, Vv, representing low complex, more or less complex and very complex respectively, and a specific score to measure the image complexity. This provides great flexibility since users can choose the form they want based on image processing. Finally, we apply the propose image complexity evaluation algorithm on two application tasks■ image enhancement and image collage.