Certain investigations on machine learning based no reference image quality assessment schemes

Abstract

In this age of rapid growth in multimedia technology there arises the need for processing millions of images. Distortion and noises were introduced in images during image acquisition, storage, transmission, restoration and reproduction. Image quality assessment (IQA) is used to estimate the degree of distortion introduced in images. Subjective IQA denotes assessment of the quality of the image by humans. Since subjective assessment is costly and time consuming and impractical for real time visual quality monitoring, controlling and integration, hence a need for reliable objective IQA method arises. Simple objective quality assessment methods are Peak signal to noise ratio (PSNR) and Mean square error (MSE) which measures the statistical information of an image. In these methods accuracy level is very low. Hence it is desired to have objective IQA methods which are effective across various types of distortion and also consistent with Human Visual System (HVS) system. Characteristics of human visual system must be taken into consideration while designing an objective measure, this will make the algorithm quite complex. Full reference image quality method uses huge memory. For assessing the quality of the image without reference with an automated system is difficult as there exists uncertainty relations between features and quality of images. In some areas reference image will not be available for assessing its quality; in those cases there arises a need for no reference image quality assessment (NRIQA) / Blind image quality assessment Machine Learning methods have better understanding of middle level image representations in recent work it shows improved performance in object recognition, image classification and segmentation tasks. Features from raw input images are learnt using the ability of Machine learning technique. Complicated mappings can be learnt by machine learning technique even with minimal domain knowledge. The automatic learning is possible by the use of this technique without any explicit programmi