FPGA-based point processing for denoising of the show-through effect in scanned document images

Abstract

Document Image Binarization is one of the image enhancement techniques and has potential applications in the field of Document Image Analysis and Recognition (DIAR). During the DIAR process, different types of noise such as uneven illumination, clutter, margin, skew, bleed-through and show-through are removed. However, the removal of show-through noise is a challenging task because it affects mostly the document image quality and readability. In this work, we have designed and developed a model-based point processing operation, which is implemented in the Altera Cyclone® V DE1-SoC for removing the show-through noise and enhancing the degraded document image quality. The Altera Cyclone® V DE1-SoC board was chosen for this study due to its performance, size compatibility, and real-time implementation, which are better than the programmable DSP chips. The Pixel/Point processing operations were implemented using the DSP builder tool, which boost-up better execution of computational operations involved in complex algorithms. To implement the proposed algorithm, we have developed the board library file for Altera Cyclone® V DE1-SoC series 5CSEMA5F31C6 board in DSP builder and tested on various document images. The calculated quality measures of the proposed algorithm tested on various document images in the FPGA board exhibited better results in terms of document image readability, quality, processing speed, and memory size. Further, the performance of the proposed algorithms was studied in different scanned document images dataset as well as with the other traditional techniques. The results of this study witnessed that the proposed model-based point processing methods performs very well than the other methods.

Keywords

• Document image binarization
• Field programmable gate array
• DSP builder
• Point processing
• Multi-level thresholding