In order to reduce the picking intensity of the robot and improve the sorting efficiency, a method for identifying and sorting incomplete/broken biscuits is proposed in this study. Based on machine vision technology, the three-photo-comparison recognition algorithm was used to solve the problem of misrecognition of incomplete/broken biscuits due to only a part of the biscuits being photographed at the edge of the photographing area. The center coordinate position of the biscuit on the moving conveyor belt was obtained via the recognition algorithm. Comparisons were made to determine whether the difference in the center coordinate of the biscuit at the time interval t was equal to the distance traveled by the conveyor belt during time t, to confirm which biscuit images obtained by three shots were from the same biscuit and obtain the image feature data of this cookie. When the obtained biscuit image feature data were different from the visual feature parameters of the standard biscuit, it was identified as a defective product, and the center coordinates of the biscuit were outputted to the parallel robot for picking. In this study, the effectiveness of the program was verified experimentally and the factors that would affect the recognition rate of conversion are experimentally examined. Through experimental verification, it was found that the program was reliable, and the accuracy of identifying incomplete/broken biscuits remained at 100%. In addition, the height of the lenss had a small effect on the recognition accuracy rate, but exerted a greater effect on the recognizable radius (showing a linear relationship). After the speed of the conveyor belt increases to 19 mm/s, the recognition accuracy decreased linearly. When the light intensity reached 400 Lux, the recognition accuracy could be maintained at 100%. The method established in this study can well identify the incomplete/broken biscuit and provide technical support for online detection and sorting of biscuits in the future.