Enhancing the Informativeness of Multi-spectral Images by means of Multimodal Image Fusion

The research is devoted to the problem of combining graphic information from sensors of various physical nature in multispectral monitoring systems using methods of multimodal image fusion. Each of the sensors of the multispectral monitoring system allows to obtain digital images of the observed scene in different ranges of electromagnetic radiation. In this paper, we consider a two-spectral monitoring system, the frst sensor of which operates in the visible range, and the second in the infrared. The main problem with multimodal image fusion is that each partial sensor of the multispectral monitoring system represents specifc characteristics of the environment (brightness, thermal or radar contrasts of objects, etc). Another, no less important problem is the different spatial resolution of sensors of different physical nature and the inconsistency of their felds of view. Therefore, the problem of effective multimodal image fusion is not trivial. As a criterion for the effectiveness of combining graphical information in a single fused image, which should contain the maximum available useful information from various sensors, the informativeness of this image is chosen. A quantitative assessment of image informativeness is proposed to be performed using an improved method based on multicriteria analysis of image parameters. Multimodal image fusion is performed using the proposed method based on the discrete wavelet transform with the formation of low-frequency wavelet coefcients of the resulting wavelet spectrum by analyzing the regression communication model between the input images from different sensors. Confrmed experimentally that the proposed method of image fusion makes it possible to synthesize more informative multispectral images than known algorithms. The application of proposed method of image fusion for monitoring objects in difcult observation conditions (smoke, fog, low illumination) allows to increase the efciency of the multispectral monitoring system and signifcantly reduce the amount of redundant information coming to the operator of the system.