Multiscale human vision simulation for image segmentation
In this paper we present a new approach in order to develop a robust segmentation algorithm talcing into account perceptual features of the Human Visual System (HVS). Many segmentation methods show some weaknesses to deal with irrelevant or irregular textured regions. Through a watershed transform, we propose to extend the well-known segmentation algorithm JSEC. weighting the J criterion with the distance between the colour of the processed pixel and the mean perceived colour at a selected scale. This is defined by the colour information perceived by the HVS when the distance from the observer to the image grows up. We combine J criterion and colour gradient and show how to improve this by removing the irrelevant detection of chaotic texture patterns. These patterns appear mainly on natural scene, in objects like grass or hairs, and on noised regions which produce unwanted texture breaks.
E. Rollo N. Richard C. Femandez-Maloigne
SIC laboratory, University of Poitiers Bvd Marie et Pierre Curie, Teleport 2, 86962 Futuroscope Chasseneuil Cedex, France
国际会议
26th Session of the CIE(国际照明委员会（CIE）第26届大会)
北京
英文
1139-1141
2007-07-04（万方平台首次上网日期，不代表论文的发表时间）