An Image Segmentation Method for Banana Leaf Disease Image with Complex Background

In view of the complexity of original banana leaf disease image collected by smart phone, an image segmentation method is proposed. It combines with the color segmentation, Ostu segmentation, which uses the minium intra-cluster or the maximum inter-cluster gray variance to segment the image, and area threshold method. Firstly, the color segmentation method is used to remove the green background according to the color characteristics of banana leaf disease images. Secondly, converting the RGB image to YUV color space after color segmentation, in which U component image is segmented by the Ostu segmentation and area threshold method to remove the non green background. Area threshold is used to eliminate the background noise. Finally, after segmented by Ostu segmentation and area threshold method, the U image is processed by an “AND” operation with each component of the original RGB image, thus obtaining a complete disease spot target image. The proposed method is used to segment the test banana leaf disease images with complex background. The results shows that it has a better segmention performance with the averaged accuracy rate over 97% and averaged error rate merely 2.3 %, which can built a solid foundation for the subsequent feature extraction and banana leaf disease recognition.