IMAGE PROCESSING METHOD BASED ON GLOBAL MOTION ESTIMATION AND DEVICE USING THE SAME

Smart overview of the Invention

The patent application describes an image processing method leveraging global motion estimation to enhance image and video quality. It involves estimating global motion parameters between a current and a reference image frame using a neural network-based model. These parameters help generate a geometric transformation matrix, which is then used to produce an output image or video. This process aims to improve image stabilization and video compression efficiency by considering global motion, such as camera movement.

Field and Related Art

The method pertains to image processing based on global motion estimation. Traditional image and video processing techniques often involve block-wise motion estimation within a limited search range. This method enhances these traditional techniques by incorporating global motion, which accounts for broader movements like camera shifts, thereby improving image stabilization and compression without expanding the search range.

Methodology

The process begins with estimating global motion parameters using a neural network model that takes current and reference image frames as input. A geometric transformation matrix is then generated by combining these parameters. This matrix is crucial in creating an output image or video, allowing for improved handling of motion effects in the final product.

Device Implementation

An electronic device implementing this method includes a camera, memory, a video codec, and a processor. The processor, executing stored instructions, uses the neural network-based global motion estimation model to derive motion parameters. These parameters are then used by the video codec or image signal processor to create a geometric transformation matrix, which aids in generating the final output video.

Technical Details

The neural network model can include various architectures like convolutional neural networks (CNNs) or deep neural networks (DNNs). The method optimizes the computational complexity by limiting the search range, crucial for devices like mobile phones with system-on-chip (SoC) constraints. By adjusting the search starting point using global motion, the method enhances block matching rates and video compression efficiency without increasing computational demands.