Abstract: Abstract: Tomato is one of the most popular and widely grown vegetables in the world. Manual harvesting of tomatoes is laborious, time-consuming and inefficient, thus making it somewhat impractical for large-scale plantations. Intelligent robots have been developed for harvesting tomato. However, as the tomato is very soft and thus especially prone to bruising, many significant technical challenges remain to be solved. In China, the research on the harvesting robot is still in its infancy, but considerable progress has been made in many aspects, such as the manipulator, image recognition, and motion control. Tomato targets extraction is the basis for location and picking of tomato. Early extraction methods have certain limitations, which are difficult to meet the demand of harvest. In this study, Niblack self-adaptive adjustment parameter selection method was put forward and successfully applied in extracting ripe tomato in greenhouse. This segmentation algorithm was based on traditional Niblack algorithm using the correlation between global and local grayscale change information of tomato image. The original tomato image was firstly transformed to gray space, and the gray-level image was obtained using the normalized color difference method, and segmented into the foreground and the background. The normalized color difference method could eliminate the light intensity information in the red and green components. Then a new Niblack threshold segmentation algorithm was used to segment the gray image. The adjustment parameter was calculated through the expected value of each window and normalized standard deviation. After denoising, the ripe tomato object could be easily extracted from segmented image by using the minimum critical rectangle method. In order to compare different segmentation algorithms, traditional Niblack algorithm, Otsu algorithm and Niblack self-adaptive adjustment parameter selection algorithm had been selected to perform the comparative analysis. Experiments showed that the Otsu algorithm could extract the target of interest in the image, which contributed significantly to the subsequent target recognition and the reduction in computation time. However, this method may fail to segment overlapping tomatoes into individual ones. For Otsu algorithm, the threshold selection in each region lacked the image characteristics, which caused the binary result to contain a lot of background noise. Traditional Niblack algorithm exaggerated image details and got a lot of unnecessary edge information, which made it difficult to separate the target from background. Niblack self-adaptive adjustment parameter selection algorithm could effectively overcome the problem of pseudo noise. This approach has gotten a good applying result in the extraction of ripe tomato object from original images in greenhouse environment. The accuracy rate of ripe tomato recognition could reach 98.3%. Compared with Otsu algorithm based on normalized difference of red and green, and traditional Niblack segmentation algorithm, segmentation algorithm based on Niblack self-adaptive adjustment parameter selection is more efficient, and its noise is smaller and the process is faster. It can meet the need of the subsequent identification of tomato image and solve the problems of low adaptation and pseudo noise block with traditional methods. But because of the complexity of the object-picking environment, the new algorithm remains to be further improved in the practical application.