基于LBP的屏幕内容图像评价CNN模型

CNN Model for Screen Content Image Assessment Based on LBP

针对传统图像质量评价模型在屏幕内容图像上存在的无法取得满意结果的问题,本文提出一种基于深度学习模型的屏幕内容图像评价模型。首先将屏幕内容图像进行归一化处理,用局部二值化(local binary pattern,LBP)算法旋转不变均匀模式求得特征图,并运用卷积神经网络对局部二值化特征图进行质量评价。为验证所提出的屏幕内容图像质量评价模型的准确性,采用斯皮尔曼秩相关系数和皮尔斯线性相关系数两种流行的评估标准进行验证。验证结果表明,本文模型与传统的质量评价模型相比具有明显的优势,表明本模型比大多数现有的图像质量评估(image quality assessment,IQA)模型更符合主观评估结果,相比于其他评价模型更具有竞争性。该研究为提升评估结果的精准度提供了理论依据。

Due to the fact that traditional image quality assessment model cannot obtain satisfactory results on the screen content image,this paper proposes a screen content image assessment model based on the deep learning model.First,the screen content image is normalized,and the local binary pattern(LBP)algorithm is used to obtain the feature map by rotating the invariant uniform pattern.The convolutional neural network is used to evaluate the quality of the local binary feature map.In order to verify the accuracy of the proposed screen content image quality assessment model,Spearman′s rank correlation coefficient and Pierce′s linear correlation coefficient are used for verification.The verification results show that the model in this paper has obvious advantages compared with traditional quality assessment models,indicating that this model is more in line with the subjective assessment results than most existing image quality assessment(IQA)methods.Compared with other models,the model is more competitive.This study provides a theoretical basis for improving the accuracy of the assessment results.