图像局部熵用于小目标检测研究

分析了局部熵用于小目标检测时造成目标范围扩散等问题的原因，并提出了熵增长方法.该方法用于点目标检测可避免发生目标范围扩散现象.由于边缘纹理和点目标在熵增长处理过程中表现出截然相反的属性，故可避免边缘纹理对于小目标检测产生的严重干扰.该方法也可用于不受噪声干扰的边缘检测.针对相同信噪比目标在不同背景亮度中具有不同熵值和增长量的问题，提出用方差增长替代熵增长，使相同信噪比目标在不同背景亮度中表现出相同的增长量值，降低了后续目标分割的难度.试验表明，熵增长方法和方差增长方法能够有效检测1或2像素大小的点目标，并不受背景中边缘纹理的干扰.对算法的复杂度进行了分析，并提出采用双通道并行流水线方式实现工程化的设计思路.

Study on Local Entropy Used in Small Target Detection

The reason for target range diffusion, posed by the local entropy when being used in small target detection, is analyzed, and an entropy rise method is proposed. The target range diffusion can be avoided by this method. At the same time, for the edge vein and point target exhibit opposite properties in the entropy rise, serious interferences in the small target detection posed by the edge vein can also be avoided, which means that the method can be used in the edge detection without noise interferences. In accordance with targets having different entropy values and entropy-rising values with the same SNR and in different background luminance, variance rise method is proposed to replace the entropy rise method. By the variance method, the same growth amounts for the targets with same SNR and in different background luminance will be exhibited and the target segmentation difficulty will be decreased. The experimental results show that the point target of 1 or 2 pixels can be detected effectively by the entropy rise method and variance rise method, and without disturb of edge veins of background. The complicacy of the two methods is analyzed, and an industrialized design using pipe-line system with dual-channel is proposed.