Blind deblurring for dermoscopy images with spatially-varying defocus blur
Yanan Lu, Fengying Xie, Zhiguo Jiang, Rusong Meng · 2016
Dermoscopy images usually suffer from spatially-varying defocus blur, which will easily influence the lesion analysis result and lead to wrong aided diagnosis. In this paper, a novel blind deblurring framework is proposed for dermoscopy images with spatially-varying defocus blur. The defocus map is firstly estimated by support vector regressor (SVR) learning model using the natural scene statistics (NSS) features. Then the blur image is divided into several focus layers based on the defocus map through a two-stage adaptive strategy. After applying a non-blind deconvolution algorithm layer by layer, the final sharp image is synthesized by the deconvolution results of these focus layers. A series of experiments show that the proposed framework can effectively restore the dermoscopy images suffering from spatially-varying defocus blur. Compared with state-of-the-art algorithms, the proposed framework delivers the best results.