Yoshikazu Iikura

Utilization of a best linear discriminant function for designing the binary decision tree

This paper describes an accurate and efficient method for supervised classification of multispectral images. First, a simple derivation of a best linear discriminant function (BLD) is presented through geometrical consideration on ellipses with equi-Mahalanosis distance. It is shown that the function satisfies the Minimax criterion, which implies the robutness with regard to prior probabilities. Then, the binary decision tree (BDT) is introduced in order to make the method efficient, where the BLD is utilized as a decision rule. Care is taken to prevent a decline in the classification accuracy during the process of constructing the BDT

Estimation of noise component in satellite images and its application

The median filter is known to be effective to estimate the amount of noise present in the image. The authors investigate its performance quantitatively, and it is compared with the Laplacian filter and the trimmed mean filter. The estimated variances are adjusted to give an unbiased estimate under ideal conditions with no structure in the image. It is also shown that the trimmed mean filter as well as the median filter are robust to simple line edge structures. The author also discusses the estimation of the covariance matrix of the noise component with interband correlation, which is useful in many algorithms for image processing, such as edge detection, data compression, and image enhancement.

Pyramid linking for satellite image classification

Proposes the use of pyramid linking as a preprocessing method for supervised classification of satellite images. Pyramid linking is a simple and efficient algorithm not only for spatial segmentation but also for data compression. As the supervised classification can be applied to the compressed data that holds the averaged values of pixels in the segment, it is expected to improve accuracy as well as computational efficiency. Using a dataset developed by TOKAI University for the verification of classification methods, performances of the proposed method are investigated. It is shown that the method is effective when the number of classes is large as in the actual classification situation.

Pyramid linking as a satellite image compression

Proposes the use of pyramid linking as an image compression method that makes much use of correlation among neighbourhood pixels as well as among multispectral bands of satellite images. The method is compared to JPEG method in terms of the compression rate and a statistical accuracy criterion. The criterion is newly defined as the Mahalanobis distance based on the covariance matrix of noise components estimated from the second spatial derivatives of the concerned image. Analysis of a LANDSAT TM image shows that the performance of the proposed method is better than the JPEG method.<<ETX>>