A neural-based approach for impoving the magnification of degraded document images from mobile terminals

Optical reading of degraded and low-resolution text requires the application of an effective super-resolution approach to improve the recognition rate. In this context, we propose a new approach to improve the finesse and reduce the complexity of an existing non-linear super-resolution method, which is inefficient on low-resolution degraded texts from mobile terminals. In order to preserve the readability of fine line text on this type of images, it is important to increase the local neighborhood perimeter without increasing computational complexity. Our approach is a new way of using of a neural architecture based on multi-layer perceptron. It represents a new technique of using a neural architecture by learning, linearizing and extending a limited super-resolution algorithm, non-linear, complex and inefficient on degraded character images. The obtained results demonstrate, also, the effectiveness of our approach compared to conventional approaches based on a direct neuronal learning of low / high resolution transformation.