An Approach of Video Quality Assessment for Video Processing System

Abstract

Video quality is the one of the characteristic to assess the video. It is measured either newlinesubjectively or objectively. In our study we used subjective video quality assessment. Our newlinemodels predict the video quality with the help of video quality metrics. It is great challenge to newlineidentify good classifier which perform better for our given dataset. newlineIn machine leaning algorithms, improving the classification accuracy for a given dataset is a newlinefundamental challenge. Machine learning algorithms have several challenges such as newlineoverfitting, class imbalance, low classification accuracy for the given datasets. Feature newlineselection techniques are adopted to improve the performance machine learning models. We newlineused Entropy and information gain techniques to select the features. Regression model is newlinedeveloped to indicate the model performance. Similarly, we adopted few filter and wrapper newlinemethods for feature selection and compared the performance of the model using state of art newlinemodels. newlineMonte Carlo model is used to find the relative importance of the features. In this study, we newlineadopted this technique to select the features. It boost the performance of a given model newlinefor video quality assessment. Performance is indicated with the help of classification accuracy newlinemeasure. Results of the proposed method show higher classification accuracy as compared newlineto traditional classification models. newlineFeature relevance is a potential domain to assign the features into particular groups. In our newlinestudy, Frequent itemsets are used to build the model. It is based on the feature relevance newlineand mainly depends on the domain knowledge. Frequent Item Mining technique is used to newlinediscover the associations between the features. It indicates correlation between the items. In newlineour study, maximum size frequent itemsets are used to develop the models. These frequent newlineitemsets are obtained from apriori algorithm and used to create the models. The proposed newlinemethod shows better classification accuracy. Superiority of the method is evaluated using newlineclassification