Research And Application Of The Intelligent Algorithms In Image Quality Assessment

AbstractThe advent of the Internet age witnesses the rapid development of the digital graphic processing techniques. In many occasions where the graphic information techniques are applied, image quality assessment is required. The image quality assessment can be divided into the subjective one and objective one. The ultimate goal of the study on the objective image quality assessment model is to design and promote an efficient assessment model with the same effect as the subjective one in replace of the subjective experiment. However, the conventional image quality assessment shows its weakness in the image information engineering. And it is becoming more evident that the result of the subjective image quality assessment is not consistent with, or even contradictory to that of the objective one. This study will conduct image quality assessment and prediction by using machine learning and extreme learning algorithm based on the comparison of the traditional technique and other techniques.As a new machine learning method, the extreme learning machine has a direct algorithm, which only needs to solve a minimum norm least square problem. Unlike other gradient-based learning algorithms, which adjust and refresh through iterative repetition, it is a supervisory learning algorithm targeted at single-hidden layer feedforward neural networks. Therefore, this kind of algorithm is characterized by less training parameter and faster speed. In contrast, conventional extreme learning machine and quite a number of improved algorithms, based on the empirical risk minimization principle, take the advantage of the minimum training error, but have the problem of overfitting, which will hinder the expansion of the models. In addition, there is much space for their improvement in the generalization performance, prediction precision and network stability. Based on ELM with the minimum empirical risk, this study will find the solution for the algorithm by utilizing LS-SVM regression learning method under the structural risk minimization principle and propose an improved M-ELM as well. It will also introduce the weighted extreme learning machine, which has better generalization performance and can handle imbalanced data.As Ada Boost algorithm can improve the prediction precision of any weak predictor, this study will combine these two algorithms by taking the above improved extreme learning machine as the weak predictor of Ada Boost algorithm. It will also come up a prediction algorithm with Ada Boost-M-WELM so as to improve the prediction precision and apply this algorithm to the experiment of the three databases in the image quality assessment. Compared with other algorithms and their combination with Ada Boost, the improved prediction algorithm proposed in this study can effectively improve the prediction precision of the image quality assessment.