Automatic object segmentation using active shape models with enhancing feature salience

Object segmentation plays a very important role for interpreting images. We present a novel method to improve feature salience of ASM for automatic object segmentation in color images. Instead of modeling local appearance on RGB color space, we construct a fisher transformed space for each feature, where contour of correspondent feature is enhanced while most surrounding contours are suppressed. Edge orientation information is chosen for representing local appearance. The experiments show that our method leads to more accurate and reliable object segmentation compared with the other color-extended ASM schemes.