A.V. Gaydenkov, E.V. Sharovatov

Optimum detection in scene of the signal of the known form (the standard) by means of Wieners filtration, in the full possible realize in the event of spatial homogeneity of the background only. The standard is occupying small part of the scene. For realization of the optimum filtration the lumpy scenes offered segmenting of the it-s on the elementary areas (EA), the size is coordinated with of value of the local spottiness of the scene. The auto correlation function (ACF) and the spectral density (SD) calculates for each EA. SD forms to margin of spectral density. Margin SD for shaping adaptive optimum detection is offered uses. For the quantitative estimation of the possibilities of the proposed approach is realized programme of detection, segmenting under investigation scene the identical square-wave elementary areas by size from 8-8 pixels. Hereinafter on the base of the margin EA occurs the shaping a margin ACF, then margin SD. A margin ACF allows to get the quantitative feature for estimation of the degree the spatial spottiness of the scene. As a result for each EA with coordinates is formed feature of the finding, which unites in margin of the finding. A solving device analyzes a margin of the finding and comes to a conclusion about presence of the standard in the scene and gives coordinates of standard. The analysis of experimental data, which was got with output of the detection is dependencies relations signal/noise from factor of the degree the spatial spottiness of the scene, has allowed the following conclusions. An attitude of signal/noise upon optimum detection at the average is 10-12 without scale-turn of the distortion and 8-10 with distortion (30, 50 tumbling and 3%, 5% zoom) and practically does not depend on contrast of the standard and nature of the scene. Using SD, valued along scene as a whole, without segmenting on EA provides the increasing a signal/noise in 1, 5-2 times than by usual correlation method. When value of the distortion, caused by tumbling on 5-70 and scaling on 5-7%, the optimum detection loses high efficiency. The Segmenting of the scene on EA, calculation of the margin ACF and SD brings about increase of the computing expenses, but increasing hardware possibilities and possibility of the optimization of the algorithm, problem of the computing powers can be solved already today. As a whole, upon observance of the requirements of the possible scale-turn distortion and small level additive noise approach, founded on use of margin SD will allow to solve the finding problem of the overland objects on qualitative new level.