A Comparative Analysis of Image Segmentation Methods with Multivarious Background and Intensity

Abstract

Segmentation is the process of separating the object into parts that generally separates the foreground and background. The techniques proposed segmentation using different approaches and based on the problems in the background pattern and intensity of the image. There are 10 (ten) techniques used are Adaptive Threshold, Region Growing, Watershed, YCbCr-HSV, K-Means, Fuzzy C-Means, Mean Shift, Grab Cut, Skin Color HSV, and Otsu Gaussian. Spoken image using American Sign Language fingerspelling of ASL University, fingerspelling primary image and the retinal image of STARE. ASL fingerspelling is fingerspelling which is standard in sign language so that the application of these ten segmentation techniques can help maximize the application of pattern recognition. While the retinal image is used to separate the blood vessels. Measuring the quality of segmentation using the Root Mean Square Error (RMSE) and Peak Signal to Noise Ratio (PSNR), the experimental results show that all tested techniques that produce an average PSNR above 40 dB, meaning segmentation techniques work well in both datasets. In ASL fingerspelling dataset, a technique that generates the highest PSNR Skin Color whereas techniques for segmentation of vessels on the Retinal dataset namely Adaptive thresholding technique.