"... In PAGE 3: ... To compare the di erent systems apos; performances, im- ages of size 512 512 such as \Lenna quot;, \Barbara quot;, and \Goldhill quot; are tested. Table1 gives the coding per- formance for di erent type of frequency partitioning. Although an adaptive nonuniform FB will be ideal, a combination of parallel and tree-structured system (C) performs fairly well.... ..."

"... In PAGE 7: ... That is, the items most similar to a should be at the top of the list and the most dissimilar items at the bottom. The maximum likelihood scores used for classifica- tion are also included in the Table2 , so that the reader can note that calculating the similarity by comparing the state paths and by comparing the maximum likeli- hood scores produce different results. As far as we know, there have not been any tests to show which technique of calculating similarity better corresponds to that of the human hearing system.... ..."

"... In PAGE 9: ... Figure 16: Visualization of cerebral blood vessels derived from a clinical MR angiography with 149 edges. some results in Table1 . The setup time includes the pre- processing step.... In PAGE 9: ... 15-17. The last line in Table1 represents a complex anatomic tree from a corrosion cast. Due to the size of this model, the Open Inventor optimization failed which explains the low frame rate in the last row.... ..."