Mikiya Hironaga

Recovery of spectral reflectances of objects being imaged by multispectral cameras.

Acquisition of spectral information of objects being imaged through the use of sensor responses is important to reproduce color images under various illuminations. In the past several models have been proposed to recover the spectral reflectances from sensor responses. The accuracy of the spectral reflectances recovered by five different models is compared by using multispectral cameras. It is shown that the Wiener estimation that uses the noise variance estimated as proposed in IEEE Trans. Image Process.15, 1848 (2006) recovers the spectral reflectances more accurately than the others when the test samples are different from learning samples.

Recovery of spectral reflectances of imaged objects by the use of features of spectral reflectances

Recovery of spectral reflectances of objects being imaged through the use of sensor responses is important to reproduce color images under various illuminations. Although the Wiener estimation is usually used for the recovery, the recovery performance of the estimation depends on the autocorrelation matrix of the spectral reflectances and the noise present in an image acquisition system. The purpose of the present paper is to show that the Wiener estimation with the noise variance estimated by the previous proposal [IEEE Trans. Image Process. 16, 1848 (2006)] and with the autocorrelation matrix that uses the features of the spectral reflectances recovered by the previous method is very effective in greatly improving the performance.

Estimating the noise variance in an image acquisition system and its influence on the accuracy of recovered spectral reflectances

It is well known that the noise present in an image acquisition system plays important roles in solving inverse problems, such as the reconstruction of spectral reflectances of imaged objects from the sensor responses. Usually, a recovered spectral reflectance vector r^ by a matrix W is expressed by r^=Wp, where p is a sensor response vector. In this paper, the mean square errors (MSEs) between the recovered spectral reflectances with various reconstruction matrices W and actual spectral reflectances are divided into the noise independent MSE (MSEFREE) and the noise dependent MSE (MSENOISE). By dividing the MSE into two terms, the MSENOISE is defined as the estimated noise variance multiplied by the sum of the squared singular values of the matrix W. It is shown that the relation between the increase in the MSE and the MSENOISE agrees quite well with the experimental results by the multispectral camera, and that the estimated noise variances are of the same order of magnitude for various matrices W, but the increase in the MSE by the noise mainly results from the increase in the sum of the squared singular values for the unregularized reconstruction matrix W.

Noise robustness of a colorimetric evaluation model for image acquisition devices with different characterization models

Colorimetric evaluation of an image acquisition device is important for evaluating and optimizing a set of sensors. We have already proposed a colorimetric evaluation model [J. Imaging Sci. Technol. 49, 588-593 (2005)] based on the Wiener estimation. The mean square errors (MSE) between the estimated and the actual fundamental vectors by the Wiener filter and the proposed colorimetric quality (Qc) agree quite well with the proposed model and we have shown that the estimation of the system noise variance of the image acquisition system is essential for the evaluation model. In this paper, it is confirmed that the proposed model can be applied to two different reflectance recovery models, and these models provide us an easy method for estimating the proposed colorimetric quality (Qc). The influence of the system noise originates from the sampling intervals of the spectral characteristics of the sensors, the illuminations and the reflectance and the quantization error on the evaluation model are studied and it is confirmed from the experimental results that the proposed model holds even in a noisy condition.

Two Methods for Color Constancy Based on the Color Correlation Matrix

Two methods for color constancy are proposed in this paper. Both are based on the correlation matrix on the three-dimensional space of colors, red, green and blue. In the first method, the eigenvector corresponding to the largest eigenvalue is assumed to be a good estimate of the illumination color, and the influence of the illumination color is eliminated from the input image. In the second method, it is assumed that the eigenvector corresponding to the largest eigenvalue presents the color gray when the illumination is white The image under white illumination is obtained by a iteration method so as to satisfy the condition that the eigenvector corresponding to the largest eigenvalue presents the color gray. These two methods were compared to a widely used typical method for color constancy, the Gray-world method, by simulation experiments using synthesized images and real images, and the effectiveness of the proposed methods was validated by these experiments.

A color constancy model with minimum brightness variance assumption

The realization of color constancy on computer vision is important to recognize objects in varying light sources. This paper proposes a method to estimate the illuminant under the “Minimum Brightness Variance Assumption” which states that the variation of the brightness of the objects is as small as possible. In this method, the illuminant is estimated to be red when the red part of the object in the scene is bright. In detail, we define an evaluation function to calculate the variance of the brightness in the scene and we minimize the evaluation function to estimate the color of the illuminant and the color of the object. We conducted experiments with synthetic images and confirmed that the proposed method works well to reduce the influence of the illuminant for the objects in the scene.

Estimating the noise influence on recovering reflectances

The evaluation of the noise present in the image acquisition system and the influence of the noise is essential to image acquisition. However the mean square errors (MSE) is not divided into two terms, i.e., the noise independent MSE (MSEfree) and noise dependent MSE (MSEnoise) were not discussed separately before. The MSEfree depends on the spectral characteristics of a set of sensors, illuminations and reflectances of imaged objects and the MSEfree arises in the noise free case, however MSEnoise originates from the noise present image acquisition system. One of the authors (N.S.) already proposed a model to separate the MSE into the two factors and also proposed a model to estimate noise variance present in image acquisition systems. By the use of this model, we succeeded in the expression of the MSEnoise as a function of the noise variance and showed that the experimental results agreed fairly well with the expression when the Wiener estimation was used for the recovery. The present paper shows the extended expression for the influence of the system noise on the MSEnoise and the experimental results to show the trustworthiness of the expression for the regression model, Imai-Berns model and finite dimensional linear model.

An algorithm for automatic extraction of contour lines from Moire image

Noise reduction is one of significant problems for Moire image processing. The detection of true con-tour lines in a noisy Moire image often requires to rejoint contours split by noises. We developed a recursive search method and a loop controlled method which identify the direction of short segment of lines in a small area to connect and repair contour lines. Unfortunately, these two mehtods can not connect contour lines without errors. When they accept human judgements, however, they are efficient and useful enough to find contour lines because the human effort to indicate whether to connect two lines or not is very slight. With these methods we can obtain contour lines more easily, faster, more accurately than before.