A Histogram-Based Heuristic for an Adaptive Active Contours Color Image Segmentation

Abstract

The fidelity to data (external energy) term in energy-based segmentation of scalar (single channel) images requires setting scalar values defining the weights assigned to the inside and outside energy functional. These values are often determined empirically, which is a tedious and time consuming task. When it comes to color images (multi-channel), the weights become vectors, which further complicates the process of identifying the appropriate weights. In this work, a new interpretation of the weight vector is introduced. It is seen as representing the contribution of each channel in the energy functional, that is equivalent to search an optimum color space. We propose a heuristic formula for estimating the values of the weight vector. It is based on the ratio of the height to the width of the color components histograms. We have applied the proposed formulation to Piecewise Constant Vector Valued (PCVV) model of Chan and Vese in both biphase and multiphase frameworks. Results of the experiments demonstrate the advantages of the proposed model over the commonly used trial and error setting of weights and the model based on color spaces mixing.