【目的】为减小土壤图像成像环境条件差异对机器视觉识别土种精度的影响，提出对野外自然环境下采集的土壤图像可控光照增强。【方法】首先，用非对称广义高斯曲线拟合土壤图像V分量直方图，并在拟合曲线中引入目标迁移量，完成图像亮度迁移，实现土壤图像全局亮度可控增强；然后，利用全局和局部信息估计图像在局部区域上的光照权重，引入目标亮度，根据权重确定局部增量，并将局部增量叠加到原始V分量，实现基于局部增量的土壤图像亮度增强；再利用sigmoid曲线，将非对称广义高斯曲线亮度迁移获得全局亮度增强结果与基于局部增量的亮度增强结果融合，获得土壤图像的亮度可控增强。最后，根据色比不变性原理，分别对原始土壤图像R、G、B分量进行颜色校正。【结果】仿真实验表明：土壤图像从低亮度向高亮度迁移时，增强图像与目标图像在V分量上各对应像素的亮度差异均值为10.526 7，与目标图像亮度均值差异为0.245 1；从高亮度向低亮度迁移时，各对应像素的亮度差异均值为10.743 0，与目标图像亮度均值差异为0.272 1；本文所提算法在亮度控制上较其他算法具有更高的精度；主观质量评价表明，以原图亮度为基准，［-30,30］为土壤图像有效增强范围；【结论】所提方法能够实现亮度可控的土壤图像增强，算法是有效的。
［Purposes］To reduce the accuracy influence that machine vision identifies soil species under the different imaging conditions, a soil image is adjusted to approximate closely a real soil image that is captured under specific natural illumination and its brightness is rated. A controlled brightness enhancement is proposed for soil images which are collected in the field environment. ［Methods］Firstly, asymmetric generalized Gaussian curve is utilized to fit the V component histogram of the soil images, and the target migration value is introduced into the fitted curve to accomplish the brightness migration of images, which realizes the brightness controllable enhancement of soil images based on the global. Then the global and local information are applied to estimate the brightness weight of an image in the spatial region. According to the weight and a given target brightness, the local increment is determined and superimposed to the original V component of the soil image to achieve its brightness enhancement. Next, the global brightness migration result of asymmetric generalized Gaussian curve is fused with the brightness enhancement result based on the local increment by taking advantage of the sigmoid curve. Finally, in accordance with the principle of color ratio invariance, the R, G and B components of the original soil image are corrected separately. ［Results］Experiment results demonstrate that the brightness absolute difference average of corresponding pixel between the enhanced image and the target image on the V component is 10.526 7, and its arithmetic mean is 0.245 1. The above indicators are 10.743 0 and 0.272 1 respectively when weakening brightness. The proposed algorithm can control image brightness with higher accuracy than other algorithms. Subjective evaluation illustrates that the effective range of soil image brightness enhancement is ［-30,30］. ［Conclusions］The simulation result proves that proposed method is effective for controllable brightness enhancement of soil image.