Color image segmentation using edge and adaptive threshold value based on the image characteristics

We propose an image segmentation method that uses adaptive threshold values based on the image characteristics to preserve object boundaries. First of all, the proposed method quantizes an image by a vector quantization and the number of quantized colors are determined using PSNR based on the image characteristics. We obtain initial regions from the quantized image and merge initial regions into semantic regions in CIE Lab color space and RGB color space. We then examine if the segmented image of RGB color space is separated into semantic objects. If the result is not satisfactorily merged, we merge the image again in the CIE Lab color space. At each stage, we use the color difference of adjacent regions as similarity criteria. Threshold values are determined according to two means: the global mean of the color difference between an original image and its split-regions; the mean of those variations. Experiment results show that the proposed method separates an image into its main objects and those boundaries are preserved. Also, the proposed method provides better results for objective measures than the conventional method.