摘要
目前基于自然场景统计的无参考图像质量评价方法 BRISQUE算法(Blind/Referenceless Image Spatial Quality Evaluator)是这类算法的典型代表;但它仅在原始图像基础上做统计分析,且忽略了各特征间的差异性。由此提出了新的改进算法BRISQUEs,并通过三个步骤实现:将被测图像的梯度图做去均值对比归一化处理,在此基础上构造新的特征向量来评价图像质量;将BRISQUE中的关键特征进行适当加权,并对图像再次评价;平均上述两次评价来进一步降低算法的偏差。通过LIVE数据库上的实验,BRISQUEs的统计评价性能明显好于之前的无参考评价算法,也要优于多尺度结构相似度指标。
Nowadays blind/referenceless image spatial quality evaluator (BRISQUE) based on natural scene statistics is one of the state-of-the-art no-reference algorithms. But it only analyzes the original image and ignores the difference of the features constructed. Here an improved algorithm BRISQUEs is proposed and implemented by three steps. First, we apply mean subtracted contrast normalized to the gradient images and construct a new feature vector to assess quality. Second, we weight some key features of BRISQUE to improve assessment. After the two assessments obtained, a further average is made to weaken the bias from different assessments. Through the experiments on the LIVE IQA database, our approach has a remarkable performance than previous no-reference algorithms and is statistically superior to the popular multi-scale structural similarity index.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2016年第12期2903-2911,共9页
Journal of System Simulation
基金
National Natural Science Foundation of China(61227802,61379082,61100129)
关键词
无参考质量评估
自然场景统计
梯度图
关键特征
no-reference image quality assessment
natural scene statistics
gradient image
key feature