-
为了验证文中算法的可行性,在Matlab2014a,Windows7、处理器主频为 2.1 GHz,内存2G 的测试平台上运行。文中选用四张彩色图像作为实验图像,各自命名为图A、B、C、D。将文中算法与直方图均衡化、SSR算法(Single Scale Retinex)、MSRCR算法(Muti-Scale Retinex with Color Restoration)、参考文献[9]中AGCWD算法做比较。SSR算法的高斯环绕尺度为80,MSRCR的高斯环绕尺度为1 580 200;文中改进随机漂移粒子群种群数量为30,最大迭代次数为50。图2~图5分别为图A、B、C、D的实验结果,其中(a)、(b)、(c)、(d)、(e)、(f)分别代表原始图像、直方图均衡化的结果、SSR算法结果、MSRCR算法结果、参考文献[9]算法结果、文中算法结果。为了定量衡量算法的性能,利用图像信息熵、标准差和平均值作为衡量标准。图像信息熵越大,表明图像内容越丰富,标准差越大,代表图像的对比度越强,图像的平均值表明图像的亮度。表1~4分别为图A,B,C,D的实验数据。
Image B Original image HE SSR MSSCR AGCWD[9] Proposed algorithm Information Entropy 6.0087 5.5465 7.0987 7.0575 5.7685 7.1085 Standard deviation 37.4061 74.3707 44.4787 44.1266 42.8914 51.9358 Mean value 26.5739 127.6286 123.8615 123.5788 122.5611 70.8161 Table 2. Experimental data of image B
Image C Original image HE SSR MSSCR AGCWD[9] Proposed algorithm Information Entropy 6.0518 5.6417 7.2255 7.1351 6.2137 7.3454 Standard deviation 33.4897 74.3102 44.7472 44.0296 37.1188 57.5542 Mean value 25.9043 127.6572 124.1044 123.9515 123.3107 73.5227 Table 3. Experimental data of image C
Image A Original image HE SSR MSSCR AGCWD[9] Proposed algorithm Information entropy 6.9521 5.954 6 7.511 9 7.568 7 7.382 1 7.2897 Standard deviation 35.626 5 74.985 1 59.727 8 59.495 4 63.164 54.8634 Mean value 146.758 9 127.494 2 128.299 6 128.676 127.428 1 145.1314 Table 1. Experimental data of image A
从图A的实验结果和表1数据来看,五种算法均能取得较好的效果,图像对比度均得到了增强,但是文中算法整体亮度更大,图形的平均值为145.1314,与原始图像的平均值146.7589更加接近,颜色更加鲜艳和柔和。从图B和图C的实验结果来看,直方图均衡化法、SSR算法、MSSCR算法的图像颜色均出现了严重的失真,参考文献[9]算法的图像颜色失真稍微次之,而文中算法在保留颜色的同时,大大提高了图像的对比度。从表2、3中的数据可以看出,文中算法的信息熵值分别为7.1085和7.3454,均高于其他四种算法,文中标准差值分别为51.9358和57.5542,高于SSR算法、MSSCR算法和参考文献[9]算法。从图D的实验结果来看,五种算法均对图像的对比度起到了增强作用,从直观效果来看,文中算法的亮度更高,更加贴近原始图像的亮度。从表4中的数据来看,文中算法的信息熵值为7.1776,仍然最高,标准差值高于SSR算法、MSSCR算法和参考文献[9]算法,亮度值最高。综合四张图像的实验结果来看,文中算法相对其他算法具有明显的优势,适用于不同的场景,颜色失真小,对比度增强效果好。
Image D Original image HE SSR MSSCR AGCWD[9] Proposed algorithm Information Entropy 6.6255 5.8916 7.0116 7.0162 7.0215 7.1776 Standard deviation 35.1738 74.9857 53.9303 54.9177 62.9009 60.4586 Mean value 162.1123 127.6125 128.3536 128.6848 127.9938 178.8673 Table 4. Experimental data of image D
Image enhancement algorithm based on trigonometric function transformation and IRDPSO optimization
doi: 10.3788/IRLA20210709
- Received Date: 2021-09-26
- Rev Recd Date: 2021-10-08
- Publish Date: 2022-08-31
-
Key words:
- image enhancement /
- trigonometric transform function /
- Laplacian operator /
- random drift particle swarm /
- image contrast
Abstract: In the complex environment, such as cloudy days, foggy days, night, weaker light illumination and other conditions, the image has a lack of contrast, and the whole is dark. In view of this problem, an image enhancement algorithm based on trigonometric function transformation and IRDPSO is proposed. The image enhancement method mainly consists of four steps. First, the color image is converted to a gray image. Then, the contrast of the grayscale image is improved by trigonometric function transformation. Then, the image is enhanced by the Laplacian operator. Finally, a color restoration process is applied to the image. Aiming at the parameters in trigonometric function transformation and the parameter selection problem of the Laplacian operator, the improved random drift particle swarm optimization (IRDPSO) algorithm is combined with an image enhancement algorithm, the fitness function is constructed by information entropy and image standard deviation, and the parameters are optimized. The proposed algorithm is compared with four other algorithms. The experimental results show that the proposed algorithm is simple, the image information entropy is enhanced, the standard difference is large, the color distortion of the image is small and the enhancement effect is better than that of the other algorithms, and the quality and contrast of the image are improved.