HDRI

Introduction

The Task

On this website, we present the results of our project for the Processing Digital Camera Images (WS09/10) seminar at the Technical University of Berlin. Our task was to develop an application for High Dynamic Range Imaging (HDRI) and to create some HDR images with it.

Dynamic Range and its Limitations

High Dynamic Range Imaging is the answer to a common problem in today's photography: In photography, dynamic range refers to the luminance (light intensity) range found in a scene or, alternatively, the luminance range of a scene that a camera can capture in one image. The problem is, that the dynamic range of cameras is limited and considerably small. For scenes with a low dynamic range, this does not pose a problem, but scenes with a great luminance range cannot be properly captured by a camera. Some parts of the scene will always be over- or underexposed.

This can be seen in the following images of the same night scene with varying exposure times. None of the images contains details for all parts of the luminance range: The first image (exposure time = 1/6 second) is mostly underexposed, except for bright lights of the shops and windows. In the third image (4 seconds), the dark parts are captured well, but the bright shops are overexposed. The image in between (1 second) is not better, as the brightest parts are already overexposed, but the dark parts are still underexposed.

High Dynamic Range Imaging (HDRI)

There are two ways to solve this problem.

The first solution is to directly increase the dynamic range of image sensors (the parts of digital cameras that turn light into electric signals). Important factors include, among others, the noise and capacity of photo diodes (the "camera pixels") with low noise levels and high capacities resulting in a greater dynamic range. That is why DSLR cameras generally have a greater dynamic range than compact cameras: The greater image sensor provides less noise and more capacity per photo diode. But even modern DSLR cameras cannot properly capture the scene above, because their dynamic range is still too limited. Camera manufacturers are currently working on this issue.

The second solution is to take these limitations as given. This is what we did. Instead of creating an HDR image with a single shot, a series of low dynamic range (LDR) images is taken and combined to create a single HDR image. This usually involves the following steps:

1. The camera is placed on a tripod and three or more images of the scene are taken with varying exposure times. For good results, the exposure time must be the only thing that changes from image to image. In other words: Camera settings, such as focus, apperture size, ISO speed or white balance must be constant, the camera must not move and the scene, too, should preferably not change at all.
2. These LDR images are used to calculate an HDR image called High Dynamic Range Radiance Map, storing pixel intensities with floating point precision instead of short integers (0..255). Furthermore, the intensities are proportional to the real light intensities of the scene (radiance/luminance).
3. Ordinary display devices cannot directly show HDR images. Their dynamic range is limited, too. To display the image, it needs to be turned into an LDR image. This process is called Tone Mapping and there are various techniques with different properties.