Study of image reconstructions of Indonesian traffic sign using Criminisi and Hung method

Abstract

Image reconstruction of traffic signs serves in order to improve the image quality of traffic signs that are covered leaves, tree branches or pole so that it can facilitate the process of recognize or detecting traffic signs. Inpainting is one of the image restoration methods to reconstruct images that have been damaged or the removal of unwanted objects where the area to be restored is based on information around the area. Criminisi introduces an inpainting method with an exemplar-based approach that combines the structure-oriented and texture-oriented scheme. While Hung popularized the Criminisi method and combined it with contour construction (Bezier curve). The researcher tried to compare the two methods by using a number of images of traffic signs, one of them by using images with a size of 400 x 407 pixels, issuing the results of Peak Signal to Noise Ratio (PSNR), the Criminisi method gets results from PSNR 36,479 decibel (dB), while the Hung method gets results from PSNR 36,827 dB. This means the result of Hung method is more similar to original image than Criminisi method.