|
|
Yi-Fan Chou1,2, Vien Cheung2, Changjun Li3, M. Ronnier Luo2, San-Liang Lee1
|
1National Taiwan University of Science and Technology (Taiwan), 2University of Leeds (UK), and 3University of Science and Technology Liaoning (China)
Download paper
|
|
Play (17min) |
Download: MP4 | MP3
|
|
This paper evaluated four conventional methods for reflectance recovery: smoothness method [1], principle component analysis [2], basis functions with smoothness constraint [3] and Wiener estimation [4,5]. Most of these methods adopt a “learning-based” procedure with a training set. Modifications based on the training set were applied for improving the reflectance recovery performance. This paper described combined methods involving the application of localised training data and localised training data with a weighted matrix to the four recovery methods [1-5]. All these methods were applied to recover reflectance from XYZ values for two datasets. Both the training and testing performance were evaluated in terms of CIEDE2000 colour differences. The results showed that the performance of the methods with localised training data significantly improved. There are also limited improvements by applying the weighted matrix. Overall, the localised weighted method (using a local training set with a weighted matrix) with Weiner estimation method performed the best.
