A new method proposal to enhance foreground images against noisy backgrounds: Haytham Thresholding

Since only pixel intensities are taken into account in the binarization of gray images during the thresholding stage, it brings with it a significant problem. Because, since the relationship between pixels in the image is neglected, it is seen that noises are sometimes defined as an object, sometimes plays a role in changing the detected object, especially in noisy images where illumination is not uniform. In this study, a locally adaptive thresholding algorithm called Haytham Thresholding is proposed in order to eliminate these limitations of global thresholding algorithms and to eliminate noise caused by lighting during the binarization of the image. Especially in the literature, it is seen that noise is high in methods performed by taking the standard deviation into account when the image has a gradient feature. To prevent this, pixel values were normalized by taking into account the weights of the pixels in the window region instead of their standard deviation. These normalized values were added to the matrix values obtained by the average filter and then subtracted from the original image matrix. In the experiments, the proposed method was compared with Otsu and three different local thresholding algorithms by using four different image types also used in the literature. The comparison of the methods was made both visually and with image quality metrics such as PSNR and SSIM. As a result, it has been observed that the proposed method produces successful results compared to both global thresholding and local thresholding algorithms frequently used in the literature.