EVALUATION OF THE NOISE REDUCTION FILTERS FOR THERMAL IMAGING CHANNEL OF SURVEILLANCE OPTOELECTRONIC DEVICES

Abstract

Optoelectronic devices for remote sensing are becoming more common in various fields of science and technology. Taking into account significant reduction in the cost of infrared detectors, thermal imager is increasingly being introduced into the composition of such systems. It essentially increases the functionality of the entire surveillance system, and allows you to work without external lighting and in smoke. Temperature contrasts on an object are oftenly higher than contrasts in the visible range of the spectrum. But the sensitivity of infrared detectors is less than in visible range. Therefore, thermal images have a small signal-to-noise ratio. An obligatory stage of signal processing in a thermal imagers is noise reduction. A standard way to increase the signal-to-noise ratio is to use digital filtering. An obvious simple filtering method is the use of standard filters that have been developed for conventional visible images. But there is a question of the effectiveness of the use of such filters for thermal imaging. The possibility of using standard digital filters to reduce the noise of thermal images is investigated. Filters are implemented by embedded functions in the Mathcad software environment: medfilt (M), wiener2d (M), gaussconv (M), orthosmooth (M), orthosmooth5 (M), centsmooth (M). To evaluate the effectiveness of using filters based on real thermal imaging, statistical metrics MSE, PSNR, SSIM, CC were calculated. Filtered images with additive, pulsed and multiplicative types of noise are considered. Recommendations are given on the use of the considered digital filters for various types of noise. None of the considered filters is recommended to eliminate periodic noise. To eliminate the “salt and pepper” impulse noise, the median filter is most effective. To eliminate the multiplicative noise and noise distributed according to the Gauss law, it is more efficient to use a 3x3 convolution kernel filter. Comparison of filter operation can be applied when designing adaptive filters on microcontrollers to reduce image noise of a thermal imaging channel of surveillance optoelectronic devices.