Dual-Camera Zooming via Patch-Based Super-Resolution

The patent merges optical zoom and digital zoom functionalities into a hybrid system. The optical zoom module provides initial magnification with high quality, while the digital zoom module further processes the image through sophisticated algorithms including patch-based super-resolution and deep learning techniques. This combination allows mobile devices to achieve high-resolution zoomed images without requiring large optical lenses, thus maintaining compact device size while improving image quality.

The patent replaces pure mechanical/optical zoom systems with a computational approach. Instead of relying solely on physical lens movement and optical magnification, the system uses digital image processing, patch-based super-resolution algorithms, and deep learning models to achieve zoom. This substitution allows for high-resolution zooming in compact mobile devices without the size and cost constraints of traditional optical zoom systems.

The patent segments the zooming function into two distinct modules: an optical zoom module for initial magnification and a digital zoom module for further enhancement. The optical zoom module uses a fixed optical system with specific focal length, while the digital zoom module processes the captured image through computational algorithms. This segmentation allows the device to achieve high-quality zoom without requiring a single complex optical system, thus reducing overall device size and cost while maintaining image quality.

The patent uses patch-based super-resolution techniques where small patches from the optically zoomed image are extracted, processed through deep learning models, and reconstructed to create high-resolution zoomed regions. This copying and processing approach allows the system to generate high-quality detailed regions without requiring the entire optical system to be scaled up, thus maintaining compact device size while achieving high image quality.

The patent performs preliminary optical zoom to capture the target scene at a magnified scale before applying digital processing. The optical zoom module first creates a magnified image, and then the digital zoom module processes this pre-magnified image using patch-based super-resolution and deep learning techniques. This preliminary optical action provides a higher quality starting point for digital processing, ensuring that the final zoomed image maintains high measurement precision and detail accuracy even in compact devices.

The patent replaces traditional digital zoom interpolation methods with advanced computational techniques including patch-based super-resolution and deep learning models. These computational methods analyze and reconstruct image details at multiple scales, preserving fine details and avoiding the blurring associated with conventional digital zoom. This substitution maintains measurement precision while enabling compact device design.