Learning Based Facial Aesthetic Prediction Using Wrapper Feature Selection Approach

Abstract

Facial aesthetics based computational prediction has gained the interest of numerous investigators and extensively grown in diverse multi-media applications. The primary objective relies over hauling out of perception aware and discriminative features to classify facial expression. To deal this, prevailing strategies may cast of filtering and segmentation, this handles handcrafted low level feature design that may not provides aesthetic perception. Here, a novel framework has been modelled for designing discriminative specification to recognize regions of interest while recognizing subject based facial aesthetic. For performing this, dataset from online source can be used that are accessible publicly. Next, use a wrapper based filtering to avoid noise in the images, this forms mid level representation associated with region of interest. Next, segmentation is done with ROI region. For given image, predicting facial aesthetic is model by any Machine learning based classifiers. This assists in bridging the gap between descriptor responses and provides improved results. The extensive simulation has been done in MATLAB environment; this will be compared to prevailing approaches. Various performance metrics in these experiments have been provided and demonstrated in anticipated scheme. The anticipated model shows better trade off in contrary to prevailing models.