Underwater MAD Blurred Image Classification by Versatile Interpretation of Machine Algorithmic Learning Classifiers

Blur is one of the major degradation problems in underwater images and it is quite difficult to identify the type of blur in the water medium. So, a subjective evaluation is done for the identification of blur types by the frequency spectrum. Then, textural features like contrast, correlation, energy and homogeneity are considered as features. Based on the features and the spectrum of images, the dataset is classified objectively. The features are calculated for the image dataset collected at various real fields like lakes and ponds in and around Kanchipuram. Hence, the dataset consists of a huge number of real-time images, and machine learning is required for classification. So, analyzing the performance of machine learning classifiers is mandatory, so that the appropriate classifier will be implemented for all the underwater blur image datasets. Therefore, the versatile interpretation of machine algorithmic learning (VIMAL) classifier performances is analyzed. In this paper, statistical features, GLCM textural features and frequency spectrum features are considered as features for ML classification. The properties considered for underwater MAD (motion, atmospheric and defocus) blurred image classification resulted in higher efficiency. As a result, the ensemble SVM classifier attains an accuracy of 88% in estimating blur type.