Using kernel functions to estimate the probability density function for segmentation images

Abstract

This paper used nonparametric kernel functions to segment images using the thresholding method. This was done by transforming the image to grey and then estimating the probability density function from the grey image data and taking the highest value of the core function vector as the threshold limit. The study also found that the bandwidth parameter affects the segmentation process because it affects the estimate of the probability density function. The bandwidth parameter smooths the curve and brings it closer to the true curve because of its significant effect on bias and variance. The beam width parameter is known to depend on the size of the sample. Thus, it was found that when the parameter is used to determine the image size into equal rows and columns, it gives good, satisfactory results for segmented images that contain the most important areas of interest with removing unhelpful or important areas. Finally, the uniform Gaussian, cosine, triangular, and logistic Silverman kernel functions proved their efficiency in extracting all image features.