Small Targets Detection Of Low Contrast

The detection for small dim targets in image processing field has been the subject of intense investigation for years. Optical sensors are valuable for their strong survival capacity in the surveillance system, but with limited detection range. For small dim targets, detection range can be extended with efficient target detection arithmetic.This dissertation addresses new efficient and effective image processing schemes, which could be used to detect and recognize small targets with low contrast successfully.First of all, the characteristics of targets and the detection difficulties are analyzed. And four efficient detection methods are put forward in this paper. A software is designed to test the four detection methods. The test results show that the four detection methods are efficient for small dim targets. The method of neighborhood entropy is also transferred to DSP high-speed image processing system to test its practicality.1) The method of neighborhood entropy is used to detect faint target, which is combined with gray morphological method to improve detection probability.2) The ratio of fractal surface to fractal resolution is used to detect small artificial targets. This method has advantage of high detection ratio, and is insensitive to noise.3) Because the arithmetic of template match is easily influenced by the movement of targets, the efficient method of adaptive template update is put forward. Furthermore, the genetic algorithm is combined with template match to increase the match speed.4) A new method based on wavelet multi-resolution analysis and data fusion technology is presented for faint-target detection. This method adapts to the targets of different size, and its runtime is less than the runtime of fractal and template match method.