No-Reference Image Blur Assessment in the DWT Domain and Blurred Image Classification
Akihiko Yoneyama, Teruya Minamoto · 2015
We propose several new no-reference/blind image quality indices for blur assessment based on the discrete wavelet transform (DWT) and demonstrate that a given image can be classified based on blur by using these indices. Our approach relies on the sharpness, granularity, and L1-norm estimation of the given image in the DWT domains with a relatively long support width. Unlike conventional methods, the reference image is produced from the given image without computing special statistics or using unsupervised methods. Instead, we produce the reference image from the given image by using certain sharpening and denoising methods, and by adopting ideas from reference image quality measures such as the PSNR and SSIM in the DWT domain. We describe the detailed procedure of our method and show some experimental results that demonstrate the high performance.