野生动物监测光照自适应Retinex图像增强算法

针对野外白天不同光照对野生动物监测图像质量造成的影响,提出一种基于Retinex理论的光照自适应图像增强方法。该方法首先使用基于复合梯度的引导滤波估计图像照度分量,克服光照突变造成的伪光晕现象;然后提出一种基于Otus阈值的对比度自适应拉伸方法实现照度分量的校正,克服传统算法过度增强的问题;最后采用照度分量单通道图像计算反射分量图像,实现色彩的保真。该文采用50张保护区实地拍摄的野生动物监测图像为样本进行试验,结果表明,该文算法相比于MSRCR算法、双边滤波Retinex算法和引导滤波Retinex算法色调保真度平均提高81.00%、5.24%和3.58%,信息熵平均提高6.76%、6.23%和2.61%,峰值信噪比平均提高53.43%、5.36%和-2.85%,运算耗时减少-29.03%、78.51%和28.68%,证明该文算法可以有效克服传统Retinex理论算法的过增强、伪光晕现象和灰化效应,实现不同光照条件下野生动物监测图像的自适应增强。

Adaptive image enhancement algorithm for wild animal monitoring based on Retinex theory

Abstract: Wildlife monitoring images can be used to conduct accurate estimation of species diversity, quantity and inhabit attribution, offering scientific proofs for wildlife resource conservation.However, the quality and availability of acquired wildlife monitoring images were usually weakened due to different illumination variations in wild environments. To address this drawback, an adaptive image enhancement method based on Retinex theory was proposed in this paper.We utilized wildlife monitoring images collected at Saihanwula Nature Reserve in Inner Mongolia as experimental samples.These monitoring images were acquired from several field experimentsfrom 2010 to 2014 by using the infrared trigger cameras and they were classified into 4 illumination conditions, namely sufficient illumination condition, low illumination condition, shadow condition and overexposure condition. Firstly, we analyzed the pseudo halo phenomenon in illuminance component image estimation process caused by the traditional guided filter. The global smoothing factor of guided filtercan notbalance the halo elimination and image detail information preservation in illumination abruptchanging image regions. Therefore, we introduced the composite gradient of images to improve the guided filter algorithm. By calculating the composite gradient image, the local adaptive smoothing factor of the guided filter was obtained to achieve the joint optimal performances of the pseudo halo elimination and the dynamic range compression. In view of the over enhancement problem in conventional Retinex algorithm, a contrast adaptive stretching method based on Otsu threshold was then proposed to realize the correction of illumination component. By calculating the Otsu threshold in illumination component images, the estimated illumination component images could reach the optimal brightness extension effect at the threshold and realize the brightness improvement in dark image regions, and it could limit the over enhancement degree of the bright image regions.And it enhanced the adaptability of the algorithm to different illumination conditions. Lastly, in order to maintain the color information of enhanced images, the single channel illumination component of the corrected images and the 3 color channels of the original images were used to conduct separate calculation to maintain the correlation of the 3 color channels. It was validated that it did not increase the complexity of the algorithm. In order to prove the superiority of this algorithm, 50 wild animal monitoring images were selected randomly for validation. This algorithm was compared with MSRCR(multi-scale Retinex with color restoration) algorithm, bilateral filter Retinex algorithm and guided filter Retinex algorithm to test its quality performance of image enhancement. Compared with the other 3 algorithms, the average hue fidelity was increased by 81.00%,5.24% and 3.58%, respectively; the average information entropy was increased by 6.76%,6.23% and 2.61%, respectively; the average PSNR (peak signal to noise ratio) was improved by 53.43%,5.36% and -2.85%, respectively; the running time was reduced by -29.03%,78.51% and 28.68% respectively. Above promising results can greatly demonstrate that the proposed method is effective in addressing the over enhancement, the fogging effect and halo phenomenon in existing algorithms, and it can achieve robust illumination adaptive enhancement of wildlife monitoring images.It makes significant contribution to further automatic identification of wildlife monitoring images as well as the improvement of information and automation level in wildlife protection.