Adaptive Noise Detection Using Texture Feature Extraction and Random Forest Classification

Abstract

Image noise detection is one of the most challenging aspect in the field of image processing and analysis. Generally, the noise affects the information of image, especially when the noise level is too high. To avoid this kind of issue, image denoising schemes is developed for several applications: medical image analysis, pattern recognition, biometric authentication, etc. to reduce the noise level in the image. This experimental research concentrates on two objectives i.e. identify the noise in the corrupted image and suggest the denoising algorithm for the respective noise. To perform this operation, a new methodology is developed, initially, noise subtraction is carried-out to separate the noisy image and higher frequency image components from the digital images. Then, feature extraction is carried-out on noisy image using texture features like homogeneity, contrast, correlation, energy and two Dimension-Discrete Wavelet Transform (DWT) for extracting the optimal feature values. The obtained feature vectors are given as the input for multi-objective classifier: Random Forest (RF) in order to classify the type of noise. Experimental outcome showed that the proposed technique improved the noise prediction accuracy up to 5-20% compared to the existing methods: Neural Network (NN), Naïve Bayes (NB) classifier, K-Nearest Neighbor (KNN) and Convolutional Neural Network (CNN) by means of accuracy, sensitivity and specificity.