Estudo comparativo de técnicas de desborramento de imagem

Abstract

The low capacity of security cameras can affect an forensic investigation due to motion blur issues or lack of focus. To solve blurring problems, classic blurring techniques or those using artificial intelligence can be used. The objective of this work is to compare the two approaches, contributing to image analysis. The Wiener Filter and the Richardson-Lucy Algorithm were implemented, and an Adversary Neural Network was trained to deblur the images. The comparison of the two approaches used two quality metrics, Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The Neural Network in all tested parameters, obtained the best averages of SSIM and PSNR, ranging on average the similarity from 0.6786 to 0.86024, compared to the other two techniques. Comparing only the image processing techniques, Richardson-Lucy had the best performance averages, with SSIM averages ranging from 0.1405 to 0.5642 in relation to Wiener Filter, which ranged from 0.081 to 0, 1251.