CAJ | 학술논문

鉴于生物视觉特征对于图像的良好表征能力，提出了一种基于生物视觉特征的无参考型图像质量评价方法。对生物视觉 ST 模型进行了研究和分析，完成了对图像的稀疏化表示；利用最小二乘支持向量机回归方法训练生物视觉特征到图像质量的映射关系，获得能够预测图像质量的回归器；通过学习的回归器完成了对图像质量的评价。基于 LIVE 图像库的实验结果表明，该方法对于特定失真和交叉失真的预测误差分别为2％和5％左右，并且与目前技术条件下的质量评价方法相比具有很好的精确性和单调性。
감우생물시각특정대우도상적량호표정능력，제출료일충기우생물시각특정적무삼고형도상질량평개방법。대생물시각 ST 모형진행료연구화분석，완성료대도상적희소화표시；이용최소이승지지향량궤회귀방법훈련생물시각특정도도상질량적영사관계，획득능구예측도상질량적회귀기；통과학습적회귀기완성료대도상질량적평개。기우 LIVE 도상고적실험결과표명，해방법대우특정실진화교차실진적예측오차분별위2％화5％좌우，병차여목전기술조건하적질량평개방법상비구유흔호적정학성화단조성。
As the biological vision features show superior performance to images representation,a non-reference image quality assessment approach based on biological vision features is proposed.The standard model of biological vision is studied and analyzed,and the sparse representation of image is accomplished through the model.The mapping correlation between biological vision features and image quality scores is trained with the regression technique of least square-support vector machine,which gains the regressor that can predict the image quality.The score of image quality assessment is accom-plished with the trained regressor at last.The experimental results based on LIVE database show that the proposed approach has predicting error of 2% and 5% for specific distortion and cross-validation distortion respectively,and exhibits a superior accuracy and monotonicity compared to state-of-the-art quality assessment approaches.