Analysis of Image Processing Methods in the Context of a Basis for Recognizing Small Objects. Image processing methods for object recognition

The article analyzes the existing methods of detecting small objects from images in the channels of technical vision against a background of noise. It is shown that the method of detecting images of small objects based on the calculation of the likelihood ratio using the estimation of the mathematical expectation of samples of spatially subband vectors and their covariance matrices is promising for further research. To solve the problem of detecting small objects from images of technical vision channels, as well as to prepare data for subsequent stages (recognition and identification), a subband analysis method based on the use of new basic functions is proposed as the main tool. An experimental assessment of the quality of detection of small-sized objects using the above-described method has been carried out, showing that acceptable indicators of the probability of correct detection (0.95) with a false alarm probability of 10^−4 are achieved with a signal-to-noise ratio of more than 14. Based on the fact that noise in images is not always statically independent and additive, the assessment of the influence of spatial spectral characteristics of noise is subject to further investigation. The analysis of the influence of a statistically independent additive noise process on the quality of detection and cognition is carried out. In this case, a set of source images containing images of small objects of the type of unmanned aerial vehicles was used. To search for an object in the analyzed image, a reference image of the object was used. In the course of the study, it was found that the family of dependencies of the probability of correct detection on the signal-to-noise ratio, taking into account the given probability of a false alarm when exposed to additive white noise, is a classic type characteristic of object detection algorithms. The necessary signal-to-noise ratio has been identified, which makes it possible to achieve an acceptable probability of correct detection. The results will allow us to form a way to compare the detection quality of small objects in images for various detection algorithms. Keywords: photo and video images, methods and algorithms for synthesis, analysis and improvement of image quality, machine vision, subband methods.