Search Algorithms and Recognition of Small Details and Fine Structures of Images in Computer Vision Systems

Abstract

Search algorithms and object recognition in digital images are used in various systems of technical vision. Such systems include: vision systems of robots, the system of recognition and identification of fingerprint, authentication system for stamp on a document and many others. Description of known search algorithms and the recognition of objects in images is well represented in the literature, for example, in (Gonzalez & Woods, 2002) and (Pratt, 2001). Review of the literature shows that in most cases, problems of recognition take into account such characteristics of the object as its geometric shape and distribution of luminosity over the entire area of the object. As a criterion of recognition the standard deviation is commonly used. Spectral characteristics, the numerical moments, color characteristics, segmentation, etc. are used in addition to the basic attributes. Wavelet analysis and fractal recognition are the latest methods (Potapov et al., 2008) in image processing and pattern recognition. Algorithms for searching small details and fine structures are used in detection and analysis of the quality of images. The accuracy of search and recognition of fine details is affected by the distortions arising during the digital compression and transmission of image signals through a noisy communication channel with interference. The peak signal-to-noise ratio (PSNR) is considered nowadays the most popular criterion of noisy images (Pratt, 2001). According to this criterion the normalized root-mean-square deviation of color coordinates is calculated and the averaging is carried out at all pixels of the image. Thus, the closer the noisy image to the original, the bigger the PSNR value and therefore the better its quality we have. However this and other similar metrics (e.g., MSE) allow for estimating only rootmean-square difference between images, therefore the best results from the metrics point of view are not always correspond to the best visual perception. For instance, the noisy image containing fine details with low contrast can have high PSNR value even when the details are not visible on the background noise. A number of leading firms suggest hardware and software for the objective analysis of dynamic image quality. For example, Tektronix – PQA 300 analyzer, Snell & Wilcox – Mosalina software, Pixelmetrix – DVStation device (Glasman, 2004). Principles of image quality estimation in these devices are various. For example, PQA 300 analyzer measures image quality on algorithm of “Just Noticeable Difference – JND”, developed by Sarnoff