摘要
为了解决奶牛夜间图像存在颜色失真、边缘细节丢失与噪声干扰等低质量问题,试验提出了一种基于高斯引导滤波的彩色恢复多尺度Retinex(multi-scale Retinex with color restoration, MSRCR)图像增强算法,即先利用多尺度高斯滤波函数处理原始夜间图像得到粗糙照度分量,再利用引导滤波(guided filtering, GF)函数处理获得精确照度分量,然后结合四方向Sobel边缘检测器进行反射分量自适应权值优化,最后通过对数加法将照度分量和反射分量合成增强图像。试验以3组(1组为均匀光照灰度图像、2组为均匀光照彩色图像、3组为非均匀光照彩色图像)夜间不同成像条件下的180幅图像为研究对象,以平均梯度(mean gradient, MG)、标准差(standard deviation, S)、信息熵(information entropy, IE)、边缘保持指数(edge preserving index, EPI)、结构相似性指数(structural similarity index, SSIM)、峰值信噪比(peak signal to noise ratio, PSNR)和运行时间为评价指标,比较分析本算法和直方图均衡(histogram equalization, HE)算法、基于GF的多尺度Retinex算法(即GF+MSR算法)、MSRCR算法和光照自适应Retinex (illuminated adaptation Retinex, IAR)算法的图像增强效果。结果表明:与原图像比,本算法增强后图像的MG值较HE、GF+MSR、MSRCR和IAR算法增强后图像分别提高了72.22%、150.00%、41.67%和31.67%,IE值分别提高了30.42%、56.90%、43.28%、8.65%;S值较GF+MSR、MSRCR和IAR算法增强后图像分别提高了68.61%、74.28%、54.62%,较HE算法增强后图像降低了55.54%。本算法增强后图像的EPI值较HE、GF+MSR、MSRCR和IAR算法分别提高了34.70%、23.72%、51.41%、12.11%,SSIM分别提高了14.47%、9.33%、5.09%、18.19%;PSNR值较经HE、GF+MSR、MSRCR算法增强后图像分别提高了30.12%、11.52%、20.05%,较IAR算法增强后图像降低了0.67%;同时本算法运行时间分别较HE、GF+MSR、IAR算法减少了67.55%、32.81%、10.14%,而较MSRCR算法增加了139.47%。说明本算法能够有效增强夜间图像的边缘信息并抑制噪声干扰,提高了夜间奶牛图像的质量。
In order to solve the low-quality problems of color distortion,loss of edge details,and noise interference in nighttime images of cows,the a multi-scale Retinex with color restoration(MSRCR)image enhancement algorithm on Gaussian guided filter was proposed in the experiment.Firstly,the rough illumination component was obtained by processing the original night image with multi-scale Gaussian filter function.Secondly,the accurate illumination component was obtained by guided filtering(GF)function.Thirdly,the adaptive weight optimization of the reflection component was carried out with the aid of the four-direction Sobel edge detector.Finally,the illuminance component and the reflection component were combined into the enhanced image by logarithmic addition.The experiment took 180 images from 3 groups(namely,one group of uniform illumination grayscale image,one group of uniform illumination color image,and one group of non-uniform illumination color image)under different imaging conditions at night as the research object,and took mean gradient(MG),standard deviation(S),information entropy(IE),edge preserving index(EPI),structural similarity index(SSIM),peak signal to noise ratio(PSNR),and running time as evaluation indexes.The image enhancement effect of the algorithm and the HE,GF+MSR,MSRCR and IAR algorithms was compared and analyzed.The results showed that compared with the original image,the MG value of the enhanced image by this algorithm had been improved by 39.44%,62.22%,24.21%and 17.86%,respectively,when compared with the enhanced imaged by HE,GF+MSR,MSRCR、IAR algorithms compared with the original image;IE was increased by 30.42%,56.90%,43.28%and 8.65%,respectively;S-value was increased by 68.61%,74.28%and 54.62%,respectively,compared with that by GF+MSR,MSRCR and IAR,however,decreased by 55.54%compared with the image enhanced by HE algorithm;EPI was increased by 34.70%,23.72%,51.41%and 12.11%,respectively;SSIM was improved by 14.47%,9.33%,5.09%and 18.19%,respectively;PSNR was improved by 30.12%,11.52%,20.05%compared with that of the enhanced images by HE,GF+MSR,and MSRCR,respectively,and was decreased by 0.67%compared with the image enhanced by IAR algorithm.At the same time,the running time of this algorithm was reduced by 67.55%,32.81%and 10.14%compared with that of HE,GF+MSR and IAR,respectively,and was increased by 139.47%compared with MSRCR algorithm.The results indicated that the algorithm could effectively enhance the edge information of the night image,suppress noise interference,and improve the quality of night cow images.
作者
岳帅
秦立峰
YUE Shuai;QIN Lifeng(College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling 712100,China;Key Laboratory of Agricultural Internet of Things,Ministry of Agriculture and Rural Affairs,Yangling 712100,China)
出处
《黑龙江畜牧兽医》
CAS
北大核心
2024年第8期37-45,共9页
Heilongjiang Animal Science And veterinary Medicine
基金
国家重点研发计划项目(2019YFD1002401)
陕西省技术创新引导专项(2022QFY11-02)。
