Multi-Level Image Segmentation Using Hybridization of Whale Optimization Algorithm and Otsu’s Method over Degraded Dataset Using CNN

Basu Dev Shivahare and S.K. Gupta

Images segmentation of an image is the process of dividing the image into multiple regions. It is a preprocessing stage of computer vision. To determine the optimal set of thresholds is one of the most stimulating tasks in the field of image segmentation. The paper discusses the segmentation results of hybridization of whale optimization algorithm and Otsu's method on the degraded images. The dataset is degraded with Gaussian noise. The proposed method works on degraded dataset. The image denoising is followed by image segmentation to process the degraded images. The image segmentation process cannot be applied directly on the degraded images since they are noisy. Firstly the degraded images are denoised using convolutional neural network and later the denoised dataset is multi-level segmented in multiple regions using hybridization of whale optimization algorithm and Otsu’s method depending up on the optimal set of thresholds. The segmentation is performed using whale optimization algorithm and in whale optimization algorithm, the fitness function is calculated using Otsu’s method. The experimental testing is performed at various segmentation levels. The paper discusses the results at 3, 5, 7 and 9 levels. The analysis is performed qualitatively (based on visual quality) and quantitatively (based on parametric values). The comparative analysis is also performed by taking various conventional and non-conventional methods. The experimental testing confirms that the proposed method using otsu’s method has the caliber to show better results on degraded dataset also as compare to the original dataset.

Volume 12 | 04-Special Issue

Pages: 174-182

DOI: 10.5373/JARDCS/V12SP4/20201479