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

Table 3

The DRSN-GAN algorithm.

---

Step 1. The training dataset is selected and preprocessed. The nearest neighbor image interpolation method is used to transform the resolution of the GoPro images IB from 1280 × 720 to 256 × 256.

Step 2. The generated image is obtained by using the generator composed of DRSN.

Step 3. Input the generated image and its corresponding clear image into the discriminator to obtain the probability that the generated image belongs to the clear image.

Step 4. Train the discriminator with the probability obtained in Step 3 and judge whether the number of times of discriminator training reaches the preset number (five times). If yes, execute Step 5; otherwise, return to Step 3.

Step 5. The probability of the generated image belonging to the clear image is obtained by using the trained discriminator.

Step 6. Determine whether the generator and the discriminator reach Nash equilibrium. If so, execute Step 8; otherwise, execute Step 7.

Step 7. Train the generator with the probability obtained in Step 5, reset the number of times of discriminator training to zero, and return to Step 2

Step 8. Use the trained generation network to remove image motion blur.

---