A new automatic thresholding algorithm for unimodal gray -level distribution images by using the gray gradient information

Automatic thresholding is an important image-processing technique that extracts target regions of interest from background based on the gray-level information. There are many methods that have been proposed to provide an appropriate threshold value for an image with an evident bimodal gray-level histogram. Nevertheless, most of these methods can not accurately and efficiently segment an image with a unimodal gray-level distribution. In this study, a new approach of determining the optimal threshold is proposed to address the above issues by introducing the gray gradient of pixels. The gray gradient value of each pixel is calculated by performing an element multiplication with the image patch and the Sobel operator masks. Then, the gray gradient distribution is obtained on the basis of the results of statistical averaging the gray gradient values of the pixels with different gray-levels. By analysing the morphology of the gray gradient distribution histogram, the optimal threshold for image segmentation is obtained. Furthermore, several test images with unimodal gray-level distributions are segmented by other algorithms, such as the Otsu algorithm, MaxEntropy algorithm and Valley-Emphasis algorithm, to assess the performance of our proposed algorithm. In addition, conventional physical mercury intrusion porosimetry (MIP) and nuclear magnetic resonance (NMR) are used to test the porosity of the imaged samples, to further verify the accuracy of the new algorithm. The comparison results indicate that the proposed algorithm demonstrates a better segmentation performance over those of the other methods for an image with a unimodal gray-level distribution and results in a porosity that is in accordance with that of the MIP and NMR test.