CS129 / Project 5 / HDR

Example of a recovered high dynamic range image




    1. Introduction:

    In this assignment, we are going to implement the algorithm to recover high dynamic range images from multiple exposure time images.

    Modern camera are unable to capture the full dynamic range of commonly encountered real-world scenes, even the best possible photograph will be partially under or over-exposed.

    The method here are separated into 2 parts: recover radiance map from multiple exposure images and use tone mapping method to display the image.


    2. Algorithm:

    2.1 Recovering a radiance map from a collection of images

    The algorithm here is to adjust the response curves of the multiple exposure time images into one smooth response curve. The result can show the ratio relationships between the radiance in the real world. The tricky part here is how to construct the smooth part of the curve, we use the g''(x) = 0 to represent the smooth.


    2.2 Converting the radiance map into a display image

    The algorithms to convert the radiance map into a display image can be different. The global one is simple, we can use log(L), sqrt(L), or L/(L+1) to adjust the range. I use the combination of Reinhart Operator and gama correction to implement the global tone mapping.

    The other one, which will produce better results, is local tone mapping. The main part of the algorithm is to use bilateral filter to separate the large scale intensity and detail intensity. Reduce the contrast in large scale and add detail, we can achieve the edge preserved image, which is much better than merely using filter on pixels.



    3. Extra Credit:

    I implemented the fast bilateral filter using Picewise-linear bilateral filtering and sub-sampling, the result seems to be related with sample rate. The intensity weight part was implemented as a 3-dimensional matrix, in which the third dimension represents the index of segmentation. The result is not as good as the original algorithm, too much sample and approximation.

Results in a table

The Response Curve

The Detail intensity

The Combined Intensity

Local Tone Mapping

Global Scale

Global simple