基于客观参数的图像质量评估

Image Quality Assessment Using Objective Parameters Based on Wavelet Decomposition

图像质量评价采用小波变换把自然图像分解成十个子带,在每个子带中设置两个参数和两个Logistic函数。通过计算客观度量与主观评估测试结果之间的关系来定义两个基于感知的模型,评估测试在包含344幅JPEG和JPEG2000图像的数据库中进行。线性预测模型能够反映图像的真实主观评分,而相关预测模型则能用来区分图像间的相对质量。在这两个模型中,每个客观参数和主观感知特性的关系用两个Logistic函数逼近,从而获得每个参数的最佳估计缺损程度。最终的图像质量度量通过组合各个估计缺损程度获得。实验表明这两种模型性能优良,和主观感知的图像质量有很强的相关性。

The paper proposes a method for image quality assessment using objective parameters based on wavelet decomposition. Natural images are decomposed into ten subbands using wavelet transform. In each subband, two parameters and two logistic functions are assigned. Two perception - based models that can predict the subjective perception are defined by computing the relationship between objective measures and the results of subjective assessments tests, and applied to a database which contains 344 JPEG and JPEG2000 images. The linear prediction model can be used to mirror the real subjective scores of the images and the correlation prediction model can distinguish the relative quality between/among images. In these two models, the relationship between each objective parameter and the sub- jective perception is approximated by two logistic functions, resulting a best estimated impairment level for each parameter. The final image quality metric is obtained by the combination of the estimated impairment levels. Experiments show that the two models proposed provide excellent performance , and the subjective perception of the image quality is very relevant.