基于暗通道先验和区间插值小波变换的图像去雾霾方法

针对在雾霾天气条件下采集到的图像质量退化,影响现代化农业精准作业的问题,该文提出一种基于暗通道先验理论和区间插值小波变换的图像去雾新方法。该文将暗通道先验模型和区间插值小波变换相结合,期望能有效滤除雾霾信息,恢复景物颜色特征,使图像更加清晰。结果表明:经过该方法处理后,图像整体较明亮,图像的对比度和清晰度都得到提高,达到滤除图像中雾霾的效果。主观上符合人眼的观察感受,图像的层次感突出,景物细节纹理也保持较好,彩色图像的色彩饱和度被很好地保持住,图像的失真度较低,逼近景物的真实颜色。去雾效果与暗通道先验算法对比,该文算法标准差数值在R通道平均提高25.44%;G通道平均提高27.90%;B通道平均提高26.24%。因此,采用该方法可以实现图像去雾,为进一步准确获取图像信息奠定基础,适应于现代精准农业的应用。

Image dehazing method based on dark channel prior and interval interpolation wavelet transform

Abstract: Nowadays, smart agriculture has become a research hotspot in the field of agriculture technology. Meanwhile, the image is one of the important data sources for smart agriculture and related technology. Image processing technology has been widely used in modern agricultural research. In the application of outdoor agriculture, environmental conditions are important factors that degrade the quality of the obtained image. In particular, the haze is a very common factor that decreases image quality seriously. Images acquired in bad weather, such as haze, are seriously degraded by the scatting of the atmosphere particles, which reduces the contrast, color saturation and hue shift and makes the object features difficult to identify. In order to remove the negative effect of the haze in degrading image quality, this study proposes a new image dehazing algorithm that combines the dark channel prior model with the interval interpolating wavelet transform. The dark channel prior is based on the statistics of the haze-free outdoor images. Specifically, it is based on a key observation, i.e. most local patches in haze-free outdoor images contain some pixels which have very low intensities in at least one color channel. Using this prior with the haze imaging model, we can directly estimate the thickness of the haze and recover a high quality haze-free image. The wavelet transform is used to carry out multi-scale refinement through the operation of telescopic translation, which can highlight the characteristics of the details of the image. Interval interpolation wavelet may reduce the error caused by the approximation of the wavelet. Firstly, we estimate the transmission and the atmospheric light value by using the dark channel prior theory, and restore the image. Secondly, the obtained image is decomposed by interval interpolation wavelet transform, and then reconstructed by processing high frequency sub-band wavelet coefficients. An experiment is carried out using this method. The results show that after the image processing by using this method, the whole image looks like comparatively bright, and the image contrast and clarity are improved. Finally, it works to filter out the negative effect caused by the haze. The processed image fits the human observation feeling well. It has good visual effect, obvious layering and rich texture detail. For color images, color saturation can be well kept, and the distortion is correspondingly low. Hence, the processed color images are close to the real objects with true color. Moreover, after the image processing, the contour contrast of the scene is obvious and is not blurred. It also makes distant scenery in the image very clear. We compare our haze removal results with that by the dark channel prior algorithm. On average, the standard deviation values of our algorithm in the R, G, and B channels are respectively improved by 25.44%, 27.90% and 26.24%. In sum, this study presents a new method that combines the dark channel prior model with the interval interpolating wavelet transform, and the image can be well dehazed and achieve good restoration in image visibility using this method, and thereby lays the foundation for acquiring accurate image information. Moreover, it is also useful for the application in modern precision agriculture.