Segmentations techniques based on background subtraction and color models: a comparative survey

Segmentations Techniques based on Background Subtraction and Color Models: A Comparative Survey
Bruno Brandoli Machado, Wesley Nunes Gon¸alves, Jonathan de Andrade c Silva, Vin´ ıcius Assis Saueia, Kleber Padovani de Souza, and Hemerson Pistori
Dom Bosco Catholic University Research Group in Engineering and Computing Tamandar´ Avenue, 6000, Jardim Semin´rio, 79117-900, Campo Grande, MS, Brazil e a {bmachado,wnunes,jsilva,vsaueia,kleber}@acad.ucdb.br {pistori}@ucdb.br http://www.gpec.ucdb.br

Abstract. This paper presents the comparation between two image segmentation groups, one of them based on background subtraction and the other based in color model supervised learning. The comparations were realized based on real images, manually segmented, referring two important problens that are being studied by several computer vision researches groups: sign language interpretation and mouse behavior analysis for health experiments area. For the problem of sign language interpretation were used images containing many signers, differents skin tones, static background environments and complex, in laboratory and in external environments, under differents illumination conditions. For this problem, was evaluated the capacity of the segmentators to extract from image regions containing face and hands of a human being. The experiment with mice was based on open field evaluation, a very utilized test to behavior analysis after utilizing some droug. In this problem was the image segmentation of mouse regions in image. Key words: Image Segmentation, Background Subtraction, Color Model

1

Introduction

Segmentation is an important step on image analysis in computer vision systems. On segmentation process the image is divided into interesting and irrelevant objects. This process tries to eliminate paltry objects of the image according to the problem requirements. There is not one segmentation technique which always obtains acceptable results for all applications [1]. Thus, the