Method and apparatus to correct digital image blur due to motion of subject or imaging device

What is this patent about?

’484 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. Traditional photography often struggles with low-light conditions where slow shutter speeds are necessary, leading to motion-induced distortion. While mechanical stabilization exists, it often requires heavy and expensive lens components that can compromise optical quality.

The underlying idea behind ’484 is to replace or augment mechanical stabilization with a computational approach that reconstructs a sharp image from motion-affected data. The invention recognizes that blur is essentially a mathematical convolution of a true image with a two-dimensional transfer function representing the path of motion. By identifying this motion path—either through hardware sensors or software-based blind estimation—the system can apply an inverse operation to recover the original, undistorted scene.

The claims of ’484 focus on a method and apparatus for generating a corrected image by capturing a sequence of multiple fast-shutter images rather than a single long exposure. The system identifies a main subject within each frame of the sequence and computationally shifts these frames vertically and horizontally to align that specific subject at a consistent coordinate. These aligned frames are then merged pixel-by-pixel to produce a final two-dimensional photographic image where the subject appears sharp.

In practice, the invention functions by capturing several images back-to-back at the highest possible shutter speed to minimize the blur within any individual frame. Because these fast exposures are naturally darker and noisier, the system uses pixel-level alignment to stack the images, effectively summing the light data to achieve the exposure of a slow shutter speed without the associated motion smudge. This alignment can be driven by pattern recognition, a tracking signal from the subject, or manual user designation via a viewfinder.

This approach differs from prior art by moving the correction from the optical/mechanical domain into the digital processing domain. Instead of trying to keep the sensor perfectly still during a long exposure, the system allows for movement and corrects it by coherently combining multiple short-exposure snapshots. This allows for the recovery of a high signal-to-noise ratio while ensuring that a moving subject or a shaking camera does not result in a lost or unusable photograph.