Experimental Studies on the Recognition of Small-Sized Objects in Video Images Using Multidimensional Spatial-Subband Vectors

Abstract

Background: A decision rule has been developed to recognize small-sized objects in video im-ages. The input data for the decision rule are the samples of space-subband vectors formed from the image of objects. Experimental studies of the decisive function are carried out using images of various small-sized objects. They demonstrate obtained numerical values of the like-lihood ratio logarithm used to make a decision on object recognition. It is shown that the devel-oped decision rule allows to recognition of small-sized objects in video images when carrying out a priori training.

Methods: The use of sub-band representation of video images allows to keep both the spatial and frequency structure of the object image. This can give an advantage over other methods using the recognition rules with a preliminary description and the parameters of the recognition feature informativeness. It is possible to use the methods of subband analysis and signal syn-thesis.

Results: The studies were carried out using images of unmanned aerial vehicles of the copter type and showed an image of an object according to which a training sample (a reference im-age) and a stud-ied image were formed. Experimental studies showed that the largest values of the likelihood ratio logarithm were located in proportion to those pixels of the image under study, on which the object is located.

Conclusion: The produced decision rule authorizes to recognition of various small-sized ob-jects in video images with high-quality indicators. The invented approach to the construction of the decision rule makes it possible to use optimal solutions and the Neumann-Pearson criterion to set the threshold level.