Automatic Color Image Segmentation Based on Illumination Invariant and Superpixelization

Superpixel and invariant methods for color images are becoming increasingly popular in many applications of computer vision and image analysis. This paper presents an automatic segmentation based on illumination invariant and superpixelization methods. We develop an automatic superpixel generation method by automatically modifying the quick-shift parameters based on invariant images. The proposed method segments a color image into homogeneous regions by applying quick-shift method with initial parameters, and then automatically get the final segmented image by calculating the best similarity between the output image and the invariant image by changing the quick-shift parameters values. To reduce the number of colors in image that will be used in comparison, a quantization process is applied to the original invariant image. Changing parameters values in iterations instead of using a specific value made the proposed algorithm flexible and robust against different image characteristics. The effectiveness of the proposed method for a variety of images including different objects of metals and dielectrics are examined in experiments.