Complexity Reduction in Intra Prediction of HEVC Using a Modified Convolutional Neural Network Model Incorporating Depth Map and RGB Texture

The use of videos has become widespread across various domains, including personal, entertainment, business, and education. As the demand for video storage and transmission continues to grow, the need for efficient compression techniques becomes crucial. Raw video data occupies a significant amount of memory storage, resulting in higher costs. To address this challenge, compression standards have been developed, but they often come with computational complexity issues. Attempts to reduce complexity can lead to decreased video quality and frequent buffering. Nowadays, neural networks and deep learning approaches have emerged as powerful solutions in various fields. They offer promising solutions to tackle complex problems effectively. In the case of video compression, the Coding Unit (CU), which is the fundamental block in the latest High Efficiency Video Coding (HEVC) standard, undergoes quadtree splitting for prediction purposes. However, this splitting process significantly contributes to complexity. To overcome this challenge, a convolutional neural network (CNN) model Codec with multimodal Alex Net was proposed to predict the split of the quadtree. The novel approach has proved to reduce time complexity by − 66.92% compared to other standard algorithms, while also keeping the bitrate increase to a minimum. By employing the capabilities of deep learning, a significant reduction in computational complexity was achieved without compromising video quality. The results demonstrate the effectiveness of the proposed approach in addressing the complexity issues associated with the quadtree split in HEVC.