Video inpainting using hybrid DMN-EfficientNet with LGSR

Video inpainting attempts to take off a specific area of a video or object and replace it in a visually consistent way. Identifying any interesting object or event has become a very time-consuming task because there is far more recorded video than the operators can watch. To overcome this, the video inpainting approach, Deep Maxout Network-EfficientNet (DMN-EfficientNet) is proposed in this paper. The input videos are subjected to video frame extraction. Then, residual frame extraction is performed using the Deep Residual Network (DRN). The extracted residual frames are then subjected to initial level restoration using Low-rankness guided Group Sparse Representation (LGSR). The output generated and the extracted residual frames are used as the input to the DMN-EfficientNet for video inpainting. The DMN-EfficientNet is devised using Deep Maxout Network (DMN) and EfficientNet. The performance of DMN-EfficientNet is estimated with metrics, like Second Derivative Measure of Enhancement (SDME), Structural Similarity (SSIM) and Peak Signal to Noise Ratio (PSNR), and the values attained are 78.07, 0.986 and 40.19 dB, respectively.