HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Yoshioka, Y. Articles by Ohsawa, R. Search for Related Content PubMed Articles by Yoshioka, Y. Articles by Ohsawa, R. Agricola Articles by Yoshioka, Y. Articles by Ohsawa, R. Related Collections Rice Seed Quality Data acquisition and assimilation

CROP ECOLOGY, MANAGEMENT & QUALITY-NOTES

Chalkiness in Rice: Potential for Evaluation with Image Analysis

^a Graduate School of Life and Environmental Sciences, Univ. of Tsukuba, Tsukuba, Ibaraki 305-8572, Japan, present address: National Institute of Vegetable and Tea Science, National Agriculture and Food Research Organization, Tsu, Mie 514-2392, Japan
^b Graduate School of Life and Environmental Sciences, Univ. of Tsukuba, Tsukuba, Ibaraki 305-8572, Japan
^c National Agricultural Research Center, National Agriculture and Food Research Organization, Tsukuba, Ibaraki 305-8666, Japan
^d Plant Biotechnology Institute, Ibaraki Agricultural Center, Mito, Ibaraki 311-4203, Japan

^* Corresponding author (osawaryo{at}sakura.cc.tsukuba.ac.jp).

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Chalkiness is a major concern in rice (Oryza sativa L.) breeding because it is one of the key factors in determining quality and price. Evaluation of chalkiness is traditionally performed by human visual inspection, and there is no standard objective method to effectively classify chalky grains into different categories. In this study, we evaluated the effectiveness of image information processing with an inexpensive personal computer and a digital image scanner to measure and categorize chalkiness and assessed the method's viability as an alternative to human visual assessment. A support vector machine based on the image data generated an accuracy rate of 90.2% in discriminating the level of chalkiness, and principal-components analysis of the image data provided new quantitative variables related to the location and degree of chalkiness with much greater accuracy than was previously possible. These results indicate that image processing may be a useful tool for evaluating the chalkiness of rice in scientific research and breeding programs.

Abbreviations: DFT, discrete Fourier transform • FFT, fast Fourier transform • LOOCV, leave-one-out cross-validation • PC, principal component • PCA, principal-components analysis • RBF, radial basis function • SVM, support vector machine

Chalkiness in Rice: Potential for Evaluation with Image Analysis

^a Graduate School of Life and Environmental Sciences, Univ. of Tsukuba, Tsukuba, Ibaraki 305-8572, Japan, present address: National Institute of Vegetable and Tea Science, National Agriculture and Food Research Organization, Tsu, Mie 514-2392, Japan
^b Graduate School of Life and Environmental Sciences, Univ. of Tsukuba, Tsukuba, Ibaraki 305-8572, Japan
^c National Agricultural Research Center, National Agriculture and Food Research Organization, Tsukuba, Ibaraki 305-8666, Japan
^d Plant Biotechnology Institute, Ibaraki Agricultural Center, Mito, Ibaraki 311-4203, Japan

^* Corresponding author (osawaryo{at}sakura.cc.tsukuba.ac.jp).

Chalkiness is a major concern in rice (Oryza sativa L.) breeding because it is one of the key factors in determining quality and price. Evaluation of chalkiness is traditionally performed by human visual inspection, and there is no standard objective method to effectively classify chalky grains into different categories. In this study, we evaluated the effectiveness of image information processing with an inexpensive personal computer and a digital image scanner to measure and categorize chalkiness and assessed the method's viability as an alternative to human visual assessment. A support vector machine based on the image data generated an accuracy rate of 90.2% in discriminating the level of chalkiness, and principal-components analysis of the image data provided new quantitative variables related to the location and degree of chalkiness with much greater accuracy than was previously possible. These results indicate that image processing may be a useful tool for evaluating the chalkiness of rice in scientific research and breeding programs.

Abbreviations: DFT, discrete Fourier transform • FFT, fast Fourier transform • LOOCV, leave-one-out cross-validation • PC, principal component • PCA, principal-components analysis • RBF, radial basis function • SVM, support vector machine

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
DEGRADATION OF GRAIN quality, such as abnormal morphology and coloration, occurs frequently in rice (Oryza sativa L.), owing to high temperatures during the ripening period. Unfavorably warm weather at the grain filling stage results in reduced enzymatic activity related to grain filling, respiratory consumption of assimilation products, disturbed water balance, and decreased sink activity of glumous flowers (Inaba and Sato, 1976; Sato and Inaba, 1976a,b). These physiological disorders are major factors that prevent the progress of normal grain filling. Among several grain qualities affected by these physiological disorders, chalkiness is a major concern in rice breeding because it is one of the key factors in determining rice quality and price. If part of the milled rice grain is opaque rather than translucent, it is characterized as chalky. Chalkiness disappears on cooking and has no effect on taste or aroma; however, it detracts from the appearance and thus downgrades milled rice (IRRI, 2006).

The chalky appearance is associated with the development of numerous air spaces between loosely packed starch granules and the resulting change in light reflection (Tashiro and Wardlaw, 1991). A major cause of chalkiness is considered to be exposure to high temperatures during the ripening period (Tashiro and Wardlaw, 1991). To overcome this problem, breeders are carrying out various studies to better understand the genetic and physiological mechanisms that govern chalkiness (He et al., 1999; Tan et al., 2000; Cheng et al., 2003; Patindol and Wang, 2003; Lin et al., 2005; Tabata et al., 2005; Wan et al., 2005). In Japan, chalky grains are conventionally classified into five categories based on the nature of the chalky appearance: milky-white rice, white-core rice, white-belly rice, white-based rice, and white-back rice (Fig. 1 ). In addition, grains on which the base and back parts appear chalky are sometimes classified into an additional category (white-back and -based rice). Although it is easy to classify milky-white, white-core, and white-belly rice on the basis of human visual judgment, it is difficult to effectively classify the other rice categories (white-based, white-back, and white-back and -based rice). This is because the degree and locations of chalkiness in these categories differ among cultivars, among plants within a cultivar, and even among grains within a panicle. Therefore, human visual assessment sometimes results in unacceptable errors, and training and experience are required for accurate and consistent assessment. Moreover, there is no effective method for quantifying the location and degree of chalkiness, possibly because these characteristics exhibit continuous variation among the categories. Therefore, development of an appropriate method for objective assessment of chalkiness is highly desirable.

View larger version (28K): [in this window] [in a new window]   Figure 1. Grayscale images (upper row) and binary images (lower row) of rice grains: perfect rice (PR), white-based rice (WBSR), white-back rice (WBCR), white-back and -based rice (WBBR), white-belly rice (WBR), white-core rice (WCR), and milky-white rice (MWR). Since the grains were illuminated from behind, the chalky parts of the grains appear darker than normal parts.

The objective of the present study was to develop a new method for objective classification and quantification of chalkiness by means of image information processing. Specifically, we considered whether personal computers and digital image scanners could provide an alternative to human visual assessments.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Rice Grain Samples
We used perfect and chalky grains in each chalkiness category from 12 rice cultivars: Hokuriku-11, Tarehonami, Ohu, Togo, Houkoku, Moritawase, Akitakomachi, Hanaechizen, Hitachi-21, Hatsuboshi, Sasanishiki, and Chiyonishiki. Accurate classification of grains as perfect or immature was conducted by human visual assessment using binary images (see below) according to the method of Hoshikawa (1975). A total of 246 perfect and chalky grains were used in the following analysis: perfect rice (82 grains), milky-white rice (28), white-core rice (28), white-belly rice (25), white-based rice (19), white-back rice (16), and white-back and -based rice (48).

Image Information Processing
A perfect rice grain is translucent, allowing the transmission of scattered light, whereas opaque or chalky areas in a chalky grain prevent this transmission (Tashiro and Wardlaw, 1991). Illumination from behind, in which light passes through the tissues of the rice grain (transillumination), provides clearer contrast between the chalky and nonchalky parts of the grains than is possible with illumination from the front. Although a linear-scanning camera and a high-performance digital camera can be used effectively to obtain high-resolution images, the high cost of this equipment may represent a barrier to the adoption of scanning technology. We therefore attempted to perform our image analyses with a comparatively inexpensive image scanner and personal computer. Current digital image scanners are reasonably priced, provide advanced features, and are able to scan film positives. Scanning the rice grains by using this function provided images similar to those obtained by means of transillumination. We used a CanoScan 9950FV digital image scanner (Canon, Inc., Tokyo, Japan) to obtain grain images with the scanner's film positive scanning mode. Each grain was placed on the flat bed of the scanner and scanned at 300 dpi resolution. Each image was saved in red–green–blue (RGB) color JPEG format with 256 levels (i.e., 8-bit resolution per channel). These high-quality images of the grains were converted into grayscale images (Fig. 1). To sharpen the difference between the chalky and normal parts of the grains, we also converted the grayscale images into binary images by using a threshold method. In this approach, a threshold value was calculated by subtracting the standard deviation of the grayscale value from the mean of all pixels in the perfect rice images and was used as the basis for comparison in all images of chalky grains. Any pixel with a value lower than this threshold value was defined as chalky. The resulting binary images were used in the visual classification of grains into perfect or immature categories. Before subsequent analysis, each grayscale image was rotated around its center of gravity so that the central axis of the grain corresponded approximately to a vertical line and so that the embryo was positioned at the lower left of the image (Fig. 2A ).

View larger version (21K): [in this window] [in a new window]   Figure 2. Schematic diagrams of the data-acquisition process using grayscale images of rice grains. (A) Rotation of each grain image so that the long axis was vertical and the embryo was at the bottom left of the image. Dashed contours represent cases in which the original image was rotated by 30, 60, and 90 degrees. (B) The original chalkiness curve created from the dataset containing the mean grayscale values and five reconstructed chalkiness curves based on a constant plus the first two, four, eight, and 16 Fourier coefficients. (C) Division of a rice grain into 18 sections.

Each chalkiness curve can be regarded as a discrete periodic function of the angle and grayscale values, and each dataset was thus subjected to a discrete Fourier transform (DFT), as follows. The sequence of N complex numbers x_n (n = 0, 1, 2, ..., N – 1) was transformed into a sequence of N complex numbers X_m (m = 0, 1, 2, ..., N–1) by means of DFT according to the formula

where e is Napier's constant (the base of natural logarithm) and i is the imaginary component of the number (i.e., the square root of –1). The inverse DFT is given by

The first sample, x₀, of the inversely transformed sequence is the average of the sequence X_m. In addition, when X_m is a sequence of real numbers (i.e., the imaginary part of X_m = 0), the sequence x_n is symmetric about half of the sampling frequency. In the case of our data, X_m is a dataset of mean grayscale values and N is equal to the number of grayscale values. Then, the Fourier coefficients were calculated by multiplying the real part of x_n by 0.5 and multiplying its imaginary part by –0.5 (n = 1, 2, ..., N/2). These Fourier coefficients, including the first sample x₀ (hereafter referred to as constant), were used to characterize the chalkiness curves. We referred to this method as the chalkiness curve method.

The sampling interval (in degrees) was determined so that each dataset comprised 2ⁿ values so that we could use the fast Fourier transform (FFT) technique, which is considerably faster than the DFT technique. This decision allowed us to use an inexpensive computer to perform the calculations instead of a more powerful (and correspondingly more expensive) computer, thereby lowering the barriers to adopting our analytical approach. In a preliminary study, we used four sampling intervals: 11.25° (32 mean grayscale values), 5.625° (64 values), 2.8125° (128 values), and 1.40625° (256 values). We then compared the results of the classification using each of the four intervals to classify the grains into seven rice categories. The preliminary results demonstrated that the 5.625° interval provided the best discrimination between grain categories (data not shown), indicating that this interval was able to generate chalkiness curves that adequately characterized the location and degree of chalkiness. In addition, this sampling interval also represented an acceptable compromise between image processing speed and classification accuracy. Therefore, we adopted a sampling interval of 5.625° (64 mean grayscale values) to create the chalkiness curves and computed the DFT efficiently by using the FFT algorithm of Cooley and Tukey (1965).

The second type of image data was obtained by modifying an existing method described by Akiyama et al. (1996). In this method, the rice grain is divided into four quadrants, and the proportion of the total size of each quadrant occupied by the white core is calculated. We modified this approach as follows. Since differences between categories were not detected in preliminary analyses using only four quadrants, we subdivided each grain into 18 sections (Fig. 2C) and calculated the mean grayscale value of the pixels in each section. These 18 grayscale values were then used as characteristics of the grain's chalkiness. We refer to this method as the portioning method.

Classification
We used a support vector machine (SVM), which is a widely used learning algorithm (Vapnik, 1995; Burges, 1998; Schölkopf and Smola, 2002), to categorize each grain into one of the seven rice categories. An SVM classifier can be applied to linearly or nonlinearly separable data, with or without overlap of the class data (Burges, 1998). Consider a simple binary classification problem. For training samples that are not linearly separable, the data are first transformed from the input space to a higher-dimensional feature space via a function, , such that z = (x) is the feature-point corresponding to data item x. Then, two parallel hyperplanes are determined so as to maximize the distance between them and simultaneously minimize the number of training points in the area between the planes (i.e., the margin). Finally, a third hyperplane that passes through the middle of the margin is defined. The discriminant equation for the SVM classifier is defined as

where _i are weight parameters, k(x, x_i) is the kernel function employed for the data transformation into the linearly separable feature space, x_i are the support vectors, m is the number of support vectors, x is the input pattern vector, b is the bias or threshold, and y_i{–1, +1}, depending on the class. In this study, the radial basis function (RBF) kernels, most commonly used in SVM classification problems, were employed:

where is the standard deviation. The above binary classification scheme can be extended easily to N classes, where N > 2.

To demonstrate the performance of the SVM, accurate classification rates were calculated by means of leave-one-out cross-validation (LOOCV), a technique used to evaluate the performance of a classifier building on a dataset of n data. First, the method takes n – 1 data as the training set from the original dataset. Using the SVM classifier, the value of the data point that was left out, p, is predicted., and the error between the true value of p and the prediction is obtained. In a similar way, these two steps are repeated for different datasets of n – 1 data, each with one data point "left out." Finally, the accuracy of the classification is calculated. To determine the minimum number of Fourier coefficients required to obtain a highly accurate classification with the SVM classifier for each of the seven categories, Fourier coefficients were added beginning with the coefficients for low harmonics. That is, the reconstructed chalkiness curve based on the Fourier coefficients approximated the original curve more closely as the number of Fourier coefficients increased. In addition, we also calculated the accuracy of the classification for the case with 18 mean grayscale values obtained for the 18 sections defined by the portioning method. To determine the robustness of the image-processing procedure, we compared procedures without regard to cultivar. These statistical analyses were performed by using version 1.7.1 of the R software (R Development Core Team, 2005) and its e1071 package (Dimitriadou et al., 2006).

Quantitative Measurement
To obtain quantitative characteristics related to the chalkiness curve, we performed a principal-components analysis (PCA) based on the variance–covariance matrix of the Fourier coefficients obtained by the chalkiness curve method. In the PCA, we used 10 Fourier coefficients of the first five harmonics, since our analysis revealed that these coefficients provided the best fit to the data. Yoshioka et al. (2004) previously demonstrated the effectiveness of this PCA approach for quantitatively evaluating the color and mosaic patterns of images of Primula sieboldii petals. To quantify the location and degree of chalkiness, we also performed a PCA based on the variance–covariance matrix of the 18 mean grayscale values obtained by the portioning method. The principal component (PC) scores were used as the quantitative characteristics of chalkiness in two subsequent analyses. First, ANOVA was performed to investigate the differences among the seven rice categories (perfect rice and six categories of immature rice). Second, each pair of categories was examined by using the Tukey-Kramer multiple-comparison procedure.

	RESULTS AND DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
Classification by the Support Vector Machine
The original chalkiness curve was approximated adequately by using nine Fourier coefficients (the constant plus eight coefficients; Fig. 2B). On the basis of these nine Fourier coefficients, the reconstructed chalkiness curves for perfect rice and for the six categories of immature rice grains shown in Fig. 1 correctly distinguished the different characteristics of each category of immature rice (Fig. 3 ). The maximum classification accuracy in distinguishing immature rice categories by using the LOOCV was 85.8% for the case with the first 10 Fourier coefficients (Fig. 4 ). With fewer than 10 Fourier coefficients, the rate of accurate classification increased as the number of Fourier coefficients increased. In contrast, the rate of accurate classification decreased as the number of Fourier coefficients increased beyond 10. This degradation of the rate of accurate classification may have resulted from overfitting (i.e., where the classifier adjusts to very specific features of the training data, its performance improves for the training examples but degrades for other data). This result also indicated that the characteristics of the seven rice categories were fully described by the low-frequency component of the Fourier descriptors and that the high-frequency component, which describes minute changes, was less important for characterizing the rice categories.

View larger version (15K): [in this window] [in a new window]   Figure 3. Reconstructed chalkiness curves for the perfect rice grain category and for the six immature rice grain categories shown in Fig. 1, using the first nine Fourier coefficients (including one constant). The categories are perfect rice (PR), white-based rice (WBSR), white-back rice (WBCR), white-back and -based rice (WBBR), white-belly rice (WBR), white-core rice (WCR), and milky-white rice (MWR).

View larger version (16K): [in this window] [in a new window]   Figure 4. Accuracy of the leave-one-out cross-validation (LOOCV) data in a classification problem using a support vector machine with a radial basis function (RBF) kernel based on the Fourier coefficients of the chalkiness curves for the perfect rice grain category and the six immature rice grain categories. Accuracy represents the overall rate of accurate classification. The number of Fourier coefficients was increased, in order, starting with the coefficients of the low harmonics.

Quantitative Evaluation by Means of Principal-Components Analysis
A good summary of the 10 Fourier coefficients of the chalkiness curve was provided by the first two PCs. The contributions of the first two PCs to the total variance were 81.9% and 12.2%, respectively, for a cumulative contribution of 94.1%. In the portioning method, the first two PCs provided a good summary of the 18 mean grayscale values. The contributions of the first two PCs to the total variance were 67.4% and 18.1%, respectively, for a cumulative contribution of 85.5%. Table 3 shows the variation in PC1 and PC2 in both methods for all rice categories. The orders of the mean values suggest that the first PC of both methods provided a good measure of the degree of chalkiness and that the second PC of both methods accounted for the location of the chalkiness. For example, PC1 in the chalkiness curve method produced the lowest score for milky-white rice and the highest score for perfect rice, indicating that as the degree of chalkiness increases, PC1 decreases. For PC2 in the chalkiness curve method, the grains in which chalkiness appeared in the left part of the image had the highest scores, whereas grains in which chalkiness appeared in the right part of the image had the lowest scores.

View larger version (25K): [in this window] [in a new window]   Figure 5. Variation in the first two principal components (PC1 and PC2) in perfect rice grains (PR) and in six immature rice categories: white-based rice (WBSR), white-back rice (WBCR), white-back and -based rice (WBBR), white-belly rice (WBR), white-core rice (WCR), and milky-white rice (MWR). (A) PC1 vs. PC2 for the chalkiness curve method. (B) PC1 vs. PC2 for the portioning method.

Until now, chalkiness has been measured by calculating the proportion of the total number of grains that are immature within a plant or panicle (i.e., the rate of occurrence of chalkiness). The results of our image information processing and PCA indicate that we can now quantitatively evaluate the location and degree of chalkiness with much greater accuracy than was previously possible. For grains of white-based rice, white-back rice, and white-back and -based rice, which are difficult to visually categorize into classes, the quantitative measures described in this report may be effective. Moreover, the new measures based on PC scores would be useful for genetic and physiological studies of chalkiness.

	CONCLUSIONS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
New methods for classifying chalkiness in rice were developed on the basis of image information processing. These methods require only an inexpensive commercially available computer and digital scanner. The excellent overall accuracy of the classification (85.4 and 90.2% in the chalkiness curve and portioning methods, respectively) indicates that these methods have considerable promise for application in scientific research and breeding programs. However, the low classification accuracy of rice grains in highly similar categories, such as white-based rice, white-back rice, and white-back and -based rice, indicates that further improvement of these methods will be required. Nonetheless, we successfully demonstrated the effectiveness of quantitative methods for evaluating chalkiness as an alternative to human visual categorization. Now and in the future, many researchers carrying out studies to understand better the genetic and developmental mechanisms that govern chalkiness will benefit from using these methods by being able to rapidly, inexpensively, and objectively quantify the parameter they are studying. This should eliminate inconsistent assessments resulting from inexperience of the estimator, human error, and differences among estimators. Our methods are more objective than human visual assessment, and with further development they will enable us to develop a unified methodology for the assessment of chalky grains. In cases where grain quality appears as an abnormal color pattern in other crops, this method could be used efficiently with additional research to adapt the approach to the specific characteristics of the crop.

	ACKNOWLEDGMENTS

 
This work was supported by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science.

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Received for publication October 3, 2006.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION CONCLUSIONS REFERENCES

 

• Akiyama, Y., Y. Takahara, and K. Yamamoto. 1996. The application of image analysis to valuation on white-core in kernel of rice for sake brewery. (In Japanese with English abstract.). Breeding Sci. 46:367–371.
• Akiyama, Y., H. Yamada, Y. Takahara, and K. Yamamoto. 1997. Comparison of standard images of longitudinal sections and transverse sections for evaluation of white-core in rice kernel for sake brewery. (In Japanese with English abstract.). Breeding Sci. 46:379–384.
• Burges, C.J.S. 1998. A tutorial on support vector machines for pattern recognition. Data Mining Knowledge Disc. 2:121–167.[CrossRef]
• Cheng, W., G. Zhang, G. Zhao, H. Yao, and H. Xu. 2003. Variation in rice quality of different cultivars and grain positions as affected by water management. Field Crops Res. 80:245–252.[CrossRef]
• Cooley, J.W., and J.W. Tukey. 1965. An algorithm for the machine calculation of complex Fourier series. Math. Comp. 19:297–301.[CrossRef]
• Dimitriadou, E., K. Hornik, F. Leisch, D. Meyer, and A. Weingessel. 2006. e1071: Misc. functions of Department of Statistics (e1071), TU Wien. Available at cran.r-project.org/src/contrib/Descriptions/e1071.html (verified 18 July 2007). R Foundation for Statistical Computing, Vienna, Austria.
• He, P., S.G. Li, Q. Qian, Y.Q. Ma, J.Z. Li, W.M. Wang, Y. Chen, and L.H. Zhu. 1999. Genetic analysis of rice grain quality. Theor. Appl. Genet. 98:502–508.[CrossRef][Web of Science]
• Hoshikawa, K. 1975. The growing rice plant: An anatomical monograph. Rural Culture Association, Tokyo, Japan.
• Inaba, K., and K. Sato. 1976. High temperature injury of ripening in rice plant: VI. Enzyme activities of kernel as influenced by high temperature. (In Japanese with English abstract.). Proc. Crop Sci. Soc. Jpn. 45:162–167.
• IRRI. 2006. Rice knowledge bank. Available at www.knowledgebank.irri.org/ (verified 18 July 2007). International Rice Research Institute, Los Baños, Philippines.
• Lin, S.K., M.C. Chang, Y.G. Tsai, and H.S. Lur. 2005. Proteomic analysis of the expression of proteins related to rice quality during caryopsis development and the effect of high temperature on expression. Proteomics 5:2140–2156.[CrossRef][Web of Science][Medline]
• Morita, S., J. Yonemaru, and J. Takanashi. 2005. Grain growth and endosperm cell size under high night temperatures in rice (Oryza sativa L.). Ann. Bot. 95:695–701.[Abstract/Free Full Text]
• Paige, D., S. Geng, and S. Jongkaewwattana. 1991. Apparatus for automatic measurement of kernel weight, length, and thickness. Crop Sci. 31:1314–1318.[Abstract/Free Full Text]
• Patindol, J., and Y.J. Wang. 2003. Fine structure and physicochemical properties of starches from chalky and translucent rice kernels. J. Agric. Food Chem. 51:2777–2784.[CrossRef][Web of Science][Medline]
• R Development Core Team. 2005. The R project for statistical computing. Available at www.r-project.org/ (verified 18 July 2007). R Foundation for Statistical Computing, Vienna, Austria.
• Sapirstein, H.D. 1995. Variety identification by digital image analysis. p. 91–130. In C.W. Wrigley (ed.) Identification of food-grain varieties. American Association of Cereal Chemists, Inc., St. Paul, MN.
• Sato, K., and K. Inaba. 1976a. High temperature injury of ripening in rice plant: IV. Effect of high temperature on ¹⁴C-assimilation and translocation at an early ripening period. (In Japanese with English abstract.). Proc. Crop Sci. Soc. Jpn. 45:151–155.
• Sato, K., and K. Inaba. 1976b. High temperature injury of ripening in rice plant: V. On the early decline of assimilate storing ability of grains at high temperature. (In Japanese with English abstract.). Proc. Crop Sci. Soc. Jpn. 45:156–161.
• Schölkopf, B., and A.J. Smola. 2002. Learning with kernels. MIT Press, Cambridge, MA.
• Tabata, M., Y. Iida, and R. Ohsawa. 2005. Genetic analysis of occurrence of white-back rice and basal-white rice associated with high temperature during the ripening period of rice. (In Japanese with English abstract.). Breeding Res. 7:9–15.[CrossRef]
• Tan, Y.F., Y.Z. Xing, J.X. Li, S.B. Yu, C.G. Xu, and Q. Zhang. 2000. Genetic bases of appearance quality of rice grains in Shanyou 63, an elite rice hybrid. Theor. Appl. Genet. 101:823–829.[CrossRef][Web of Science]
• Tashiro, T., and I.F. Wardlaw. 1991. The effect of high temperature on kernel dimensions and the type and occurrence of kernel damage in rice. Aust. J. Agric. Res. 42:485–496.[CrossRef][Web of Science]
• Vapnik, V. 1995. The nature of statistical learning theory. Springer-Verlag, New York.
• Wan, X.Y., J.M. Wan, J.F. Weng, L. Jiang, J.C. Bi, C.M. Wang, and H.Q. Zhai. 2005. Stability of QTLs for rice grain dimension and endosperm chalkiness characteristics across eight environments. Theor. Appl. Genet. 110:1334–1346.[CrossRef][Web of Science][Medline]
• Wan, Y.N. 2002. Kernel handling performance of an automatic grain quality inspection system. Trans. ASAE 45:369–377.[Web of Science]
• Wan, Y.N., C.M. Lin, and J.F. Chiou. 2002. Rice quality classification using an automatic grain quality inspection system. Trans. ASAE 45:379–387.[Web of Science]
• Yamada, H., Y. Akiyama, Y. Takahara, and K. Yamamoto. 1998. Effect of location and year of production on white-core in rice kernels for sake brewery evaluated by image data. Breeding Sci. 48:355–357.
• Yoshioka, Y., H. Iwata, R. Ohsawa, and S. Ninomiya. 2004. Quantitative evaluation of flower colour pattern by image analysis and principal component analysis of Primula sieboldii E. Morren. Euphytica 139:179–186.[CrossRef]

This Article Abstract Figures Only Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Yoshioka, Y. Articles by Ohsawa, R. Search for Related Content PubMed Articles by Yoshioka, Y. Articles by Ohsawa, R. Agricola Articles by Yoshioka, Y. Articles by Ohsawa, R. Related Collections Rice Seed Quality Data acquisition and assimilation