基于显著和模糊检测的浅景深作物病害图像分割

大多数现有的基于图像的作物病害诊断方法往往对输入图像的质量具有很高的要求,例如要求背景简单、大景深等等。因此这些方法的预处理过程中需要去除复杂背景,然而这个预处理较难获得理想的结果。此外,当作物病斑面积较小时,会使得获取的图像景深较浅,也导致了这些方法难以抽取精确的病斑区域。为了解决上述问题,该文提出一种利用目标检测来分割病斑图像的方法。首先,该方法对抽取的结构特征和颜色特征进行整合并对特征空间进行量化,从而得到作物病害图像的显著区域。该方法不需要进行去除复杂背景的预处理过程即可得到病斑区域的图像；同时,为了处理浅景深的病害图像,引入了模糊检测方法用以进一步过滤背景和模糊区域的图像。试验中利用多种黄瓜和水稻病害的图片,将该方法与阈值法、图切割法进行了对比,结果表明该方法在效率不明显降低时,其分割效果明显优于阈值法；在分割效果差异不大时,其运行效率明显高于图切割方法；同时,该方法能够对浅景深的作物病害图像的病斑区域进行有效的分割。

Segmentation for low depth of field crop disease images based on saliency and blurred detection

Most existing image-based methods for crop disease diagnosis usually have high requirement of the input images, including simple background, sufficient depth of field, etc.These methods always need to remove the complex background when doing image preprocessing, which lead to obtain the desired results difficultly.Besides, when the lesion areas are small, the captured micro images always have the low depth of field, which cannot be processed effectively by these methods to extract the accurate lesion areas.In order to solve these problems, the paper proposed a method which uses target detection to segment the lesion images.Firstly, by integrating the structural features and color features extraction and feature space quantization, the saliency region of crop disease images was detected.The lesion areas can be extracted without the preprocessing of removing the complex background.Meanwhile, to deal with the crop diseases images with the low depth of filed, the blurred detection was introduced to further filter the background or blurred images.The images of various diseases of cucumber and rice were used in the experiments.The experimental results showed that our method was much better than the threshold method on accuracy and much more efficient than graph cuts method on efficiency in image segmentation of the crop disease images.Meanwhile, our method can effectively extract the lesion areas from the crop disease images with the low depth of field.