Cutting tool condition monitoring technology is of great importance to improve machining efficiency and quality in automated production. Cutting tool condition monitoring based on computer vision has great advantages which traditional monitoring methods can not provide. In the paper, by means of computer vision, key techinques of tool image processing and wear condition monitoring are studied, which is of great important to guide the tool wear condition Monitoring.The characteristics of tool wear condition, wear process and standard of abrasion are analyzed. Experimental device for tool wear condition monitoring based on computer vision is established, and parameters of the key components of the device are analyzed. Taken the tool wear images abtained from the device as the object, image preprocessing, edge detecting, wear areas segmenting are carried out, then the wear values are calculated.The edges of HSS tool wear images are detected with sub-pixel accuracy inspection method based on moment invariance. The images are preprocessed with image clipping, denoising and enhancing, then the adaptive GA based Otsu threshold segment mothod is used to segment the Sobel-edge detected images. The bottom edges of the wear area are extracted with sub-pixel accuracy inspection methos based on moment invariance.The MRF model for tool flank wear image segmenting is established with the theory of Markov random field(MRF) introduced, and Iterated Conditional Mode(ICM) and Gibbs sampled algorithm are used individually to segment the images, then the results under MAP (Maximun a Posteriori) rule are obtained. It is shown that MRF image segment method is better than the traditional double threshold methods, especially, Gibbs sampled algorithm is of higher segmenting accuracy.The Multiscale Markov random field segment model is established. Using causality properties among layers of the multiscale MRF model, the model parameters are estimated with Expectation maximization(EM) algorithm, then the segment results under the criteria of Maximizer of the posteriori marginal(MPM) are obtained. The results are compared with iterative MRF ones, it is shown that Multi-MRF method is of stronger recognition ability for object region.Chain code edge searching algorithm is adopted to describe the object area edges and the appropriate threshold values of the chain code length are determined to denoise the segmented images. Then wear values VB_(max), VB_(mean) obtained with different segmenting and edge detecting methods are calculated, the comparisions of calculated results with the measured ones show that both results are in agreement well.Based on the research above, an image process system for tool wear condition monitoring is developed. With the system, image inputing , preprocessing, wear edge detecting, wear condition judgment and wear curve figuring etc. can be carried out, which is of great practical significance to improve tool wear condition monitoring based on computer vision.