Better images with HDR Capture and Low-Light Assistant (experimental research prototypes for Nokia N900)Mon, 2010-08-30 12:31
Written by Kari Pulli, Nokia Fellow and the head of the Visual Computing and Ubiquitous Imaging research team at NRC Palo Alto.
Cameras are unfortunately not as adaptable as the human eye, and in situations where there is either too much or too little light, it’s not easy, or often not even possible to capture images that would give the viewer a similar experience as the person that saw the situation with her own eyes. However, by taking several images with different image settings and then combining the input images into a single new image it is possible to capture better images.
Nokia Research Center Palo Alto has developed two new applications to address these problems. A new camera control architecture and API created jointly by Nokia and Stanford University (http://fcam.garage.maemo.org) is a key enabler for these applications.
The HDR Capture application is meant for situations with too much light, especially when the background is bright (e.g., sky) and the foreground has less light (e.g., a person). Using normal camera to capture a single image would typically lead to a black silhouette of the person with a nice sky in the background. HDR Capture measures the brightness of the scene, captures 1-3 images as the user presses the capture button, and combines them into a single image that should show details of both the bright and dark areas.
Another problematic situation is when there just isn’t enough light. If you take an image with short exposure time, the result is a noisy image. However, if you extend the exposure time, the output is often blurry. With Lowlight Assistant the camera captures two images, one short (probably noisy) and one long (probably blurry), and automatically combines the images to a new image that hopefully has the sharpness of the short exposure and the better colors of the long exposure. The user can evaluate a small area of the image to see whether the quality is sufficient. If it is, the combined image is processed in the background, otherwise the user can try again.