In order to reduce the risk of causing harm to the hands of cutting workers in the processing of frozen-fish, the detection and recognition methods for security protection system of cutting machine based on machine vision were proposed. With the workspace of cutting bandsaw under the surveillance of the color camera, the machine vision system divided the surveillance image into different dangerous regions according to the center position of the bandsaw image, and performed image processing on the acquired surveillance images. To overcome the variance of illuminance on the operation gloves of workers, the perceptual color feature recognition method was combined with the Gaussian mixture model to classify the perceptual color features and identify the workers gloves. The final evaluation for which different dangerous regions the workers hands had entered could be given after morphological image processing and statistics from the resulting images. And the machine vision system for security protection of meat cutting bandsaw machine was implemented in the laboratory. The experimental results showed that the proposed perceptual color feature with GMM could accurately identify the color features of the gloves under the variance of illuminance. By comparing with the HSV and CIE Lab color space, the proposed detection and recognition methods could achieve good robustness and have low image processing time cost, averaging at 39.18ms. And the proposed machine vision system could meet the needs of real-time and reliability for security protection system.