In order to formulate a scientific teaching resource management method, this paper proposed a research on the visual computing model of mathematical educational resources based on image processing technology. In this paper, a shared learning model based on cloud computing was constructed through HOG feature extraction, and an image denoising method was used to simplify the algorithm, and the Fast R-CNN model algorithm was improved accordingly. The improved visual computing model could directly add bounding box regression to the CNN network for training. In the experiment, the comparison test of processing speed with different number of nodes was carried out, and the performance test was carried out by applying the improved Fast R-CNN model algorithm. The experimental results showed that the 1126.4MB data packet took 36s before the improvement, the 2355.2MB data packet took 41s, the 1126.4MB data packet took 20s, and the 2355.2MB data packet took 22s. It indicated that using this improved algorithm could improve the access speed and data processing performance of the storage model.