Mixing supervised & unsupervised learning for image deblurring

We study the effectiveness of hybrid learning algorithms in the context of image deblurring. Blurring is hard to avoid in any image acquisition system and is present in all images to some extent. So, deblurring is an essential step in preparing images for higher level processing such as segmentation, recognition etc. Furthermore, mathematical modeling of the source of blurring is difficult since many factors effect the image formation process. We believe that adaptive systems with learning capabilities offer a viable solution to this problem. In our previous work, we have shown that a-committee machine with novel gating architecture produces excellent image enhancement [6, 7]. In this paper, we explore the role of the number of committee machines used. Furthermore, we propose a measuring scale to evaluate the quality of the deblurred images.