Image Motion Deblurring Based on Deep Residual Shrinkage and Generative Adversarial Networks

Table 2

The attention mechanism algorithm.

---

Step 1. Calculate the absolute value of the result after convolution.

Step 2. Obtain the feature A through global mean pooling of the input after calculating the absolute value.

Step 3. Input the feature map after global mean pooling into a small fully connected network with a sigmoid function serving as the last layer. Normalize the output between 0 and 1 to obtain the coefficient α.

Step 4. Express the threshold as α × A and use soft threshold processing to remove redundant information and noise from the input, after the absolute value is acquired in Step 1.

---