Automatic segmentation of blurry region using haar-wavelet transform

Blurry region segmentation for partial blur images is important and challenging in image processing. We proposed a blurry region segmentation algorithm using Haar-wavelet transform. Firstly, the test image is re-blurred to generate a reference image. Secondly, a 16×16-sized surrounding block for each pixel is fetched and then the block is decomposed by 3-orders Haar-wavelet transform. The ratio of the norm sum of nine horizontal, vertical and diagonal sub-components between test image and re-blurred image can represent the blurriness of the pixel. All pixel-blurriness constitute the blur map, which characterizes the distribution of image blur severity. Finally, the local blurry region is segmented according to the blur map. Experiments demonstrate that the proposed method achieves high performance scores at such indicators as precision and recall (mostly over 0.85) for both defocused and motion-blurred images. The proposed algorithm also has a high time efficiency by comparison.