Extensão do bag-of-visual-features para incorporar informação espacial na descrição de características de acordo com a percepção visual humana

Abstract

Bag-of-visual-features is a well-known approach for pattern classication and contentbased image retrieval. The entire image is described by a normalized histogram of frequencies of visual words. In this case, objects of interest on large background area or on dierent background scenes, may not be correctly recognized. In this work we propose to extract the relevant scenes of the image (foreground) according to the human being visual perception using the saliency map approach. In our proposal, the histogram is built on weight visual words according to the saliency map on two approaches, fuzzy and binary, to incorporate a distinguish between information of background and foreground. Our method was tested on databases that contain images with dierent conditions of illumination, color, rigid and scale transformations, and changes of the background. The analysis of the results demonstrates that our proposal presents signicant improvements over other tested approaches.