Download or read book Image Quality Assessment Using an Artificial Neural Network Approach written by Atidel Bouraoui and published by . This book was released on 2012 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: Image quality assessment presents a substantial interest for image services that target human observers. Indeed, Image quality can be measured in two different ways. The first, called "subjective quality assessment", is the obvious approach given the subjective nature of the visual data quality. The second one is called "objective quality assessment" that automatically allow to produce values that score image quality. There exists a large array of objective image quality assessment measures for which a taxonomic scheme has been proposed in the beginning of this manuscript. In fact, the first objective of this thesis is to provide a complete and thorough statistical predictive performance assessment of a variety of full-reference objective quality measures over number of subjectively rated image quality databases. The second is to define the image attributes that are the most relevant to its quality evaluation. Two feature selection methods have been used including the structural risk minimization and the neural network based approaches. This allowed us to develop two new objective reduced-reference image quality metrics where the image quality assessment requires the use of only a few features of the reference and the test images. The third objective of this research work is to exploit the supervised machine learning techniques, especially the multilayer perceptron based model, for automatic image quality appreciation. The system learns from the subjective quality scores and builds a model capable to further provide an objective measure that continues to match with the human opinion to any other image. The main target was to optimize the predictive performance of the developed measures according to correlation, monotonicity and accuracy. The default cost function based on error was employed for the first developed measure (that we called ECF) and a customized cost function based on correlation was proposed to design the second metric (that we called CCF). The comparative investigation to eighteen other full-reference image quality algorithms over three image quality databases shows that both ECF and CCF take into consideration the nonlinearities of the human visual system. The ECF is more accurate than the majority of the metrics under study, while the CCF outperforms all its counterparts in terms of correlation and hence monotonicity.