| 摘要 |
Disclosed in the present invention is a method for blind image quality assessment based on a conditional histogram codebook. The method comprises the following steps: extracting features of distorted images to obtain neighboring DNT (Divisive Normalization Transform) coefficient joint conditional histograms of the distorted images, wherein the distorted images comprise distorted images for training and distorted images to be assessed; connecting various curves in the neighboring DNT coefficient joint conditional histograms in an end-to-end fashion, to obtain high-dimensional feature vectors of the distorted images; clustering the high-dimensional feature vectors of all the distorted images for training in a preset database, to obtain a clustering-based conditional histogram codebook and all code words in the conditional histogram codebook; searching all the code words in the conditional histogram codebook for matched code words matched with the high-dimensional feature vectors of the distorted images to be assessed; and calculating quality assessment indexes of the distorted images to be assessed, according to the matched code words. The present invention is more delicate in depicting the degree of statistical independence of the neighboring DNT coefficient joint conditional histograms, and the calculated image quality assessment indexes are more accurate. |
