Underwater Image Quality Enhancement Using Unsharp Masking (USM) and Bacteria Colony Optimization (BCO)

Image quality enhancement is the vital part in image processing area. Not only the capturing devices or low light are responsible for low quality image but sometimes different medium, where the image is captured is a prime challenge. In case of underwater images, most of the times the images are suffer from low contrast and low sharpness problem due to presence of water. Unsharp masking (USM) is an efficient technique to improve the high frequency components originated from discrete cosine transform (DCT) process of input image as well as enhance the edge contrast to enhance the sharpness of the image for better clarity. On the other hand, classical histogram equalization (CHE) and its modified versions are very much common practice to improve the image contrast. But these techniques are suitable for air medium as well as sometimes it suffers from under enhancement, over whitening, false contouring, etc. To overcome these problem, optimization-based techniques may be applied. An optimal solution can be achieved using these optimization techniques to improve the image contrast and preserve other image characteristics. Bacteria colony optimization (BCO) is one of such techniques in the field of computational intelligence (CI) for solving such problems. An application of histogram equalization (HE)-based bacteria colony optimization (BCO) to enhance contrast of an image has been proposed in this paper. Three different image quality assessments (IQA) metrics have been used to formulate an objective function to optimize the said problem. All the underwater images have been considered to verify the potential of proposed technique. The results have been compared with different conventional and popular optimization-based techniques. The objective results have also been compared by considering various IQA metrics. The objective and visual comparison vouch the potential of proposed technique for underwater image enhancement.