Hui-Liang Shen

Reflectance reconstruction for multispectral imaging by adaptive Wiener estimation.

In multispectral imaging, Wiener estimation is widely adopted for the reconstruction of spectral reflectance. We propose an improved reflectance reconstruction method by adaptively selecting training samples for the autocorrelation matrix calculation in Wiener estimation, without a prior knowledge of the spectral information of the samples being imaged. The performance of the proposed adaptive Wiener estimation and the traditional method are compared in the cases of different channel numbers and noise levels. Experimental results show that the proposed method outperforms the traditional method in terms of both spectral and colorimetric prediction errors when the imaging channel number is 7 or less. When the imaging system consists of 11 or more channels, the color accuracy of the proposed method is slightly better than or becomes close to that of the traditional method.

Chromaticity-based separation of reflection components in a single image

The separation of diffuse and specular reflection components, or equivalently specularity removal, is required in the fields of computer vision, object recognition and image synthesis. This paper proposes a simple and effective method to separate reflections in a color image based on the error analysis of chromaticity and appropriate selection of body color for each pixel. By solving the least-squares problem of the dichromatic reflection model, reflection separation is implemented on a single pixel level, without requiring image segmentation and even local interactions between neighboring pixels. Experimental evaluation indicates that the proposed method is effective and can deal with a wide variety of images.

Simple and efficient method for specularity removal in an image

For dielectric inhomogeneous objects, the perceived reflections are the linear combinations of diffuse and specular reflection components. Specular reflection plays an important role in the fields of image analysis, pattern recognition, and scene synthesis. Several methods for the separation of the diffuse and the specular reflection components have been presented based on image segmentation or local interaction of neighboring pixels. We propose a simple and effective method for specularity removal in a single image on the level of each individual pixel. The chromaticity of diffuse reflection is approximately estimated by employing the concept of modified specular-free image, and the specular component is adjusted according to the criterion of smooth color transition along the boundary of diffuse and specular regions. Experimental results indicate that the proposed method is promising when compared with other state-of-the-art techniques, in both separation accuracy and running speed.

Improved reflectance reconstruction for multispectral imaging by combining different techniques

In multispectral imaging system, one of the most important tasks is to accurately reconstruct the spectral reflectance from system responses. We propose such a new method by combing three most frequently used techniques, i.e., wiener estimation, pseudo-inverse, and finite-dimensional modeling. The weightings of these techniques are calculated by minimizing the combined standard deviation of both spectral errors and colorimetric errors. Experimental results show that, in terms of color difference error, the performance of the proposed method is better than those of the three techniques. It is found that the simple averaging of the reflectance estimates of these three techniques can also yield good color accuracy.

Spectral characterization of a color scanner based on optimized adaptive estimation

A scanner characterization method is proposed to estimate spectral reflectance from scanner responses by using an optimized adaptive estimation method. In contrast to our previous study [J. Opt. Soc. Am. A21, 1125 (2004)], this method considers the weighting of training samples. It is demonstrated that the color accuracy of this method is only slightly affected by the number of training samples and can provide more accurate reflectance estimation.

Real-time highlight removal using intensity ratio

In this paper, we propose an efficient method to separate the diffuse and specular reflection components from a single image. The method is built on the observation that, for diffuse pixels, the intensity ratios between the maximum values and range values (maximums minus minimums) are independent of surface geometry. The specular fractions of the image pixels can then be computed by using the intensity ratio. For textured surfaces, image pixels are classified into clusters by constructing a pseudo-chromaticity space, and the intensity ratio of each cluster is robustly estimated. Unlike existing techniques, the proposed method works in a pixel-wise manner, without specular pixel identification and any local interaction. Experimental results show that the proposed method runs 4× faster than the state of the art and produces improved accuracy in specular highlight removal.

Optimal selection of representative colors for spectral reflectance reconstruction in a multispectral imaging system

In a multispectral color imaging system, the spectral reflectance of the object being imaged always needs to be accurately reconstructed by employing the training samples on specific color charts. Considering that the workload is heavy when all those color samples are used in practical applications, it is important to select only a limited number of the most representative samples. This is possible as the color charts are usually designed to cover the range of commonly imaged colors, and the color samples are redundant for spectral image reconstruction. We propose an eigenvector-based method and a virtual-imaging-based method for representative color selection by minimizing the total reflectance root-mean-squares errors. The effectiveness of the proposed methods is confirmed by experimental results when compared with existing techniques.

Estimation of spectral reflectance of object surfaces with the consideration of perceptual color space.

Due to the nonlinear transform from spectral reflectance to CIELAB values, the solution obtained by minimizing reflectance error in traditional spectral characterization methods will not be optimal when evaluated by colorimetric error. We propose a reflectance estimation method with consideration of perceptual color space. It combines two approaches, i.e., colorimetric-based spectral calculation and weighted spectral calculation. The experimental results show that the proposed method performs better than previous methods in terms of color difference with very slight degradation in spectral accuracy.

Channel selection for multispectral color imaging using binary differential evolution

In multispectral color imaging, there is a demand to select a reduced number of optimal imaging channels to simultaneously speed up the image acquisition process and keep reflectance reconstruction accuracy. In this paper, the channel selection problem is cast as the binary optimization problem, and is consequently solved using a novel binary differential evolution (DE) algorithm. In the proposed algorithm, we define the mutation operation using a differential table of swapping pairs, and deduce the trial solutions using neighboring self-crossover. In this manner, the binary DE algorithm can well adapt to the channel selection problem. The proposed algorithm is evaluated on the multispectral color imaging system on both synthetic and real data sets. It is verified that high color accuracy is achievable by only using a reduced number of channels using the proposed method. In addition, as binary DE is a global optimization algorithm in nature, it performs better than the traditional sequential channel selection algorithm.

Autofocus for multispectral camera using focus symmetry

A multispectral camera acquires spectral color images with high fidelity by splitting the light spectrum into more than three bands. Because of the shift of focal length with wavelength, the focus of each channel should be mechanically adjusted in order to obtain sharp images. Because progressive adjustment is quite time consuming, the clear focus must be determined by using a limited number of images. This paper exploits the symmetry of focus measure distribution and proposes a simple yet efficient autofocus method. The focus measures are computed using first-order image derivatives, and the focus curve is obtained by spline interpolation. The optimal focus position, which maximizes the symmetry of the focus measure distribution, is then computed according to distance metrics. The effectiveness of the proposed method is validated in the multispectral camera system, and it is also applicable to relevant imaging systems.