Detecting Interpolated Video Frames based on Convolution Neural Networks

These days, films are exchanged often and have established themselves in daily lives due to the growing number of handheld video recording devices, social media, and messaging apps. But with a variety of user-friendly video editing programs available, anyone can manipulate videos. The spread of intentionally altered videos can have serious consequences, such as the propagation of false information. Video frames interpolation is detected in this research.The proposed methodology consists of three phases. In the first phase, the video stream is divided into a series of individual frames. Preprocessing follows, which involves resizing these video frames to fixed dimensions of 256 × 256 pixels and applying a median filter to reduce noise.The median filter’s ability to eliminate the effects of noise with exceptionally high magnitudes in the input values is noteworthy. The median filter is utilized to eliminate the effects of input values that contain noise with extremely high magnitudes. The frames were interpolated using a bilinear algorithm in the second phase. Bilinear interpolation is a linear interpolation that operates in two directions, and visual distortion is reduced by using the fractional zoom computation. In the third phase, the CNN was used for eliciting the interest points from each frame and then classifying this frame as whether it is interpolated or not. CNN is capable of picking up features and classifying things automatically. The suggested technique is superior to the most advanced detectors of the video frame interpolation, according to the results, and it may also be used to locate the interpolated frames in a video. The accuracy of the proposed methodology when using a CNN classifier was 95%, while it was 70% when using an SVM classifier.