A Real Time Image Segmentation Approach for Crop Leaf

For the monitoring of crop in greenhouse with machine vision, it is difficult to extract the leaf images from natural light background, having poor imaging conditions and different plant material. To settle the problem, a color image segmentation approach consisting of fuzzy c-means clustering, morphological operation and blob analysis was proposed. Firstly, due to the low contrast in the images captured from natural light and narrow range of pixel distribute in the color space, both contrast adjusting algorithm and decor relation stretch transform were performed for image preprocess. Secondly, for the heavy computing load in the FCM, a new FCM algorithm based on color quantization (CQ-FCM) was proposed to improve the FCM computing speed greatly without affecting the segmentation result. Then, morphological filtering was performed to eliminate most noises and a two-scan algorithm was used in the connected component labeling. After that, the Blob analysis was accomplished. With the Blob analysis result, the small objects which could not be eliminated by morphological operation were filtered out and hoes in the leaf areas were filled. Finally, the leaf images were extracted. Experimental results showed that the proposed approach is effective for extracting plant leafs from complicated background and it can meet the demand of plant real-time monitoring.