Restoration and Recognition in a Loop

Abstract

In this paper we present a novel learning based method for restoring and recognizing images of digits that have been blurred using an unknown kernel.

Preprocessing of Low-Quality Handwritten Documents Using Markov Random Fields

“…These algorithms, although effective, rely on heuristic rules of spatial constraints, which are not scalable across applications. Recent research [7], [8], [20] has applied the Markov random field (MRF) to document image binarization. Although these algorithms make various assumptions applicable only to low-resolution document images, we take advantage of the ability of the MRF to model spatial constraints in the case of high-resolution handwritten documents.…”

Handwritten Carbon Form Preprocessing Based on Markov Random Field

“…It detects strokes from local statistics in different directions. In recent years, inspired by the success of Markov Random Field (MRF) in the area of image restoration [2], [3], [4], some attempts were made to apply MRF to the preprocessing of text region of degraded images [5], [6], [18]. The advantage of the MRF model over heuristics is that it can describe the probabilistic dependency of neighboring pixels or image patches, i.e., the prior probability, and learn it from training data.…”

“…We use the MRF with the same topology as adopted in [2], [3]. Different from existing MRF based algorithms for textual image preprocessing [5], [6], [18], our algorithm uses a collection of standard patches, or representatives to represent each patch of the binarized image from the test set. These representatives are obtained by clustering all patches of binarized images in the training set.…”

“…However, for 300 dpi high resolution handwritten document images, it is not feasible to learn the prior probability or energy potentials if we simply define a much larger neighborhood. Gupta et al [5], [6] studied restoration and binarization of blurred images of license plate digits. They adopted the factorized form of MRF, i.e., the product of compatibility functions [2], [3], [4].…”

Close the loop: Joint blind image restoration and recognition with sparse representation prior

“…Although much progress has been made on pure image restoration, only few works have studied the impacts of restoration on recognition, or vice versa, the effects of recognition on restoration. The method in [3] alternated between recognition and restoration to change the patch sampling prior using non-parametric belief propagation for digit recognition, with the assumption of a known image blur model. Hennings-Yeomans et al [9] proposed a method to extract features from both the low-resolution faces and their super-resolved ones within a single energy minimization framework.…”