Résumé: The task of sex determination of silkworm pupae is usually accomplished using machine learning technology. However, the captured image included much blur because of live silkworm pupae’s writhing, which makes sex discrimination more difficulty. Accordingly, the strategy to restore the blurred image is proposed in this paper. Firstly, effective sharp edges via rolling guidance filter are predicted. Then, in the kernel estimation step, L0 regularization term and the gradient prior are, respectively, to estimate the kernel and recover the intermediate latent image. Benefiting from the multi-scale computation scheme, the accurate kernel with continuity and sparsity is acquired. Finally, in the image restoration stage, a hyper-Laplacian prior is used to recover rich edges and textures in the clear image. Simulated and real tests were conducted to verify the availability of the proposed method. Furthermore, the proposed method can also be promoted to other challenging cases, including the large blur image and the image containing noises. All experiments prove the effectiveness of the proposed method.

Langue: Anglais