NEURAL NETWORKS FOR TRANSFORM COEFFICIENT RECOVERY IN VIDEO CODEC

Smart overview of the Invention

The patent application discusses a system that employs a neural network to improve video compression by recovering transform coefficients in a video codec. This system integrates with various components such as quantization, transform, and prediction circuits. The neural network aims to generate a recovered transform block with lower error rates compared to traditional inverse quantization methods. This approach is designed to enhance the quality of video compression and decompression processes.

Field of Application

The technology is positioned within the realm of video compression, particularly focusing on the use of neural networks for transform coefficient recovery. This is crucial as video consumption increases due to the proliferation of digital media platforms. The method leverages artificial neural networks (ANNs), including convolutional neural networks (CNNs), to process and enhance video data for storage and transmission efficiency.

System Components

The system comprises a sequence of components including a memory device, prediction circuitry, transform circuitry, quantization circuitry, and a neural network. These elements work together to process input blocks of images. The prediction circuitry generates predicted blocks, while the transform circuitry processes residual blocks. The quantization circuitry then quantizes these blocks, and the neural network recovers transform coefficients to minimize errors, thereby enhancing the quality of the recovered video data.

Device and Methodology

Devices incorporating this technology include a network interface, a controller, and reconstruction circuitry with a neural network. These devices receive encoded bit streams, determine decoding parameters, and utilize the neural network to generate a recovered transform block. This process aims to reduce errors more effectively than conventional inverse quantization, using a cost function to measure and minimize discrepancies.

Applications and Benefits

The disclosed technology is applicable to lossy video compression standards, beneficial for scenarios like video streaming where bandwidth is a constraint. It supports both intraframe and interframe compression, optimizing data transfer by reducing file sizes while maintaining quality. This approach enables efficient streaming with reduced latency, adapting to varying bandwidth conditions to deliver high-quality video content.