Image Quality Assessment Method Based On Digital Watermarking

Digital images will suffer from various kinds of distortion attacks when they are in the process of acquisition, processing, transmission and so on. Then the quality of the image will be impacted inevitably. It is important that how to establish an effective method to evaluate those distortion images. Thus the IQA(Image Quality Assessment) method has become one of the most hot research topics in the field of image processing. IQA method mainly contains the subjective IQA method and the objective IQA method. Image quality is determined by our eyes. However, the subjective quality evaluation method is not realistic. Therefore, designing objective image quality assessment method which is close to the subjective quality evaluation result becomes particularly important. In this paper, the RR-IQA(Reduced-reference IQA) method based on digital watermarking is discussed, and the content of this paper is organized as follows:(1) On the basis of learning typical image quality assessment in domestic and foreign, the full reference, part reference evaluation and evaluation without reference methods are discussed, and their own performance is analyzed. The principle, classification and application of digital watermark are introduced.(2) A RR-IQA metric based on zero watermarking is proposed. Firstly, image feature is extracted when the original image is sent, and the feature is saved in the QED(Quality evaluation database). Secondly, in order to make full use of the original image information, the original image is transformed by multi-level DWT(Discrete Wavelet Transform), and the low-frequency region is obtained who preserves the original image information. Because the DCT(Discrete Cosine Transform) is a nondestructive transformation, energy coefficient and the current image quality have a direct correspondence, so zero watermark information is generated based on transformed coefficient. Then do feature extraction for distorted image and the original image, and generate zero watermark of distorted image. Finally image quality is assessed based on the two zeros watermark information. Experimental results demonstrate that our proposed method has high consistency with human subjective perception on the CSIQ images database, TID2008 images database and LIVE images database I.(3) A JPEG RR-IQA method based on traditional digital watermarking is proposed. Firstly, through the contourlet transform, the low-frequency domain is saved to QED; secondly, according to the adaptive embedding strength, a digital watermark is embed into the low-frequency coefficients, and by the inversing contourlet transform, an image which is almost the same as watermarked image is obtained; then, after a watermarked image JPEG distortion, the watermark is extracted again; finally the image quality is assessed when comparing to extract the watermark and the original watermark. Experimental results indicate that this method achieves good results and has a good consistency with subjective evaluation method on LIVE images database II and TID2008 images database.