3. Speed up tone mapping algorithm

To realize the tone mapping algorithm, there are mainly two approaches. One is software implementation and the other one is hardware implementation. Hardware implementation's advantages are high speed and easy integrated ability with high dynamic range (HDR) sensor. Especially, in the HDR vedio processing applications, software implementation can hardly handle the tone mapping operator if the vedio has a decent frames per second (FPS) and resolution. In a word, low processing speed is the largest defect when software tries to realize the tone mapping algorithm. This writing introduces a novel idea to increase the speed of software implementation of tone mapping algorithm.

I think my speed up idea is suitable for many kinds of global tone mapping algorithms. So, here I choose a famous global tone mapping operator which is belong to Drago[1] to verify my idea. Drago's algorithm is based on logarithmic compression of luminance values. The details about his algorithm can be found in his paper[1]. In his paper, he also introduced a idea to speed up his tone mapping operator. And the result is very nice. However, in my eyes, his speedup idea is particular for his algorithm. It will not suitable for others' global tone mapping algorithm. And my idea is suitable for increasing many kinds of global tone mapping algorithms.

I am good at programming in Java, so the experiment is done in Java environment. In the future, I will verify it by C++. C++'s speed should be faster than Java's in this kind of applications in my idea.

The following table shows the experiment result. From the table, we can see the computational time without using speed up technology is almost three times longer than with the technology. In my idea, the speed will be faster by implementing in C++. Maybe this kind of speed up idea will help the software implementation to decrease the speed gap with the hardware real time implementation when doing the vedio processing.

In fact the speed up idea is kind of approximation operation, so we should prove the difference between before using this idea and after using this idea is very tiny. In this project, we used the parameter peak signal to noise ratio (PSNR) to prove this point. PSNR is most commonly used to measure the quality of reconstruction of lossy compression codecs (e.g., for image compression). The signal in this case is the original data, and the noise is the error introduced by compression. When comparing compression codecs, PSNR is an approximation to human perception of reconstruction quality[2]. In our case, the higher PSNR value, the closer the two images are. The following table shows the PSNR value of the above three images.

To sum up, the meaning of increasing the software implementation of tone mapping algorithm is to let it be possible to deal with the relative high speed HDR vedio processing. The details about how to increase the speed is not mentioned now, because our work has not been published now. After we published our work, I will update this writing and show more details. Thank you for your understanding.

Reference

[1]. Drago F, Myszkowski K, Annen T, et al. Adaptive logarithmic mapping for displaying high contrast scenes[C]//Computer Graphics Forum. Blackwell Publishing, Inc, 2003, 22(3): 419-426.

[2]. https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio