Method and apparatus for using motion information and image data to correct blurred images

What is this patent about?

’788 is related to the field of digital image processing, specifically addressing the problem of image blur caused by relative motion between a camera and its subject. In traditional photography, slow shutter speeds required for low-light conditions often result in motion artifacts because the camera shakes or the subject moves while the shutter is open. The patent seeks to provide a computational solution to stabilize images and video without requiring the heavy, expensive mechanical lens elements typically found in high-end optical stabilization systems.

The underlying idea behind ’788 is to treat motion blur as a mathematical distortion that can be reversed through digital manipulation of the image data. Rather than relying solely on hardware to keep the lens steady, the invention recognizes that a blurred image is essentially a convolution of a sharp image with a transfer function representing the path of motion. By determining this motion path—either through physical sensors like accelerometers or through software-based blind estimation—the system can apply a deconvolution filter to “undo” the smudge and recover the original, sharp details.

The claims of ’788 focus on a specific implementation for stabilizing video by tracking a subject across a sequence of images. The process involves detecting a specific subject within the frames, determining its precise location, and then calculating a vertical and horizontal shift measured in an integer number of pixels to compensate for movement. By shifting individual frames based on the subject's position and then combining these corrected images, the system generates a stabilized video output where the primary subject remains clear and stationary relative to the frame.

In practice, the invention allows a device to capture multiple images at high shutter speeds and then align them to create a single, high-quality result. This approach overcomes the low signal-to-noise ratio typically associated with fast shutter speeds by coherently adding the image data while the noise remains non-coherent. The system can be configured to prioritize a designated subject, such as a moving car or a person, ensuring that the specific object of interest is rendered sharply even if the background remains blurred due to the camera's movement.

This method differentiates itself from prior art by moving away from purely mechanical stabilization and simple sharpening filters, which often lose original image data. Instead, it uses pixel-level shifting and mathematical inversion to reconstruct the image. By integrating motion sensor data with digital image processing, the ’788 patent provides a way to achieve professional-grade stabilization in compact digital devices, allowing for clear captures in challenging environments where traditional steady-handed photography is difficult.