Soft computing-based non-linear discriminate classifier for multimedia image quality enhancement

Abstract

Digital image quality measurement is a significant part in image processing systems, especially for efficient transmission. In image compression, the blocking effect is generated due to co-efficient quantification which causes poor quality on the decompressed image. The learning smooth pattern transformation with Greedy method suggests high accuracy in the approximation and classification of data but transformation-invariant coding and classification of digital images were not achieved. Residual component analysis using a Bayesian algorithm estimates the parameters involved to yield a confidence interval but nonlinearity detection was not sensed effectively. In order to overcome the transformation-invariant classification defects and nonlinearity detection, a hybrid approach is introduced in digital images. The hybrid approach is a non-linear discriminate classifier for digital image quality enhancement (NDC-DQE). The NDC-DQE uses class information and finds a set of projection vectors on digital images thereby minimizing the computational complexity. Initially, NDC identifies the degradation type contained in a digital image then measure the quality of that image by using image quality measure for that specific degradation. The NDC-DQE is applied to focus on particular artifacts such as blocking effects and non-blocking effects. The NDC-DQE also uses the statistical framework that covers a large set of common degradations. The NDC-DQE with information fidelity criterion is integrated to derive the mutual information for one subband and multiple subbands. Based on, NDC covers a very large set of possible degradations usually determined in practical digital image used applications. Simulations conducted with NDC-DQE show the performance improvement in image quality in classification accuracy, peak signal to noise ratio, computational complexity, and image quality rate.