| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To implement highly accurate and highly robust image conversion that enables the relaxation of the input condition of an image that is a conversion source, reduce the processing load, increase the processing speed, and reduce the required amount of memory. <P>SOLUTION: An eigenprojection matrix #14 is generated from a learning image group #10, in which high quality images and low-quality images are paired up, by a projection operation #12 using a locality relationship, and a projection nuclear tensor #16 that defines the correspondence relationship between the low-quality image and an intermediate eigenspace and the correspondence relationship between the high quality image and the intermediate eigenspace is created. A first sub nuclear tensor is created (#24) by first setting from the projection nuclear tensor, and a coefficient vector in the intermediate eigenspace is calculated by projecting #30 an input low-quality image #20 using the eigenprojection matrix and the first sub nuclear tensor. A high quality image #36 is obtained by projecting #34 the coefficient vector using a second sub nuclear tensor #26 created by second setting from the projection nuclear tensor, and the eigenprojection matrix. <P>COPYRIGHT: (C)2011,JPO&INPIT</p> |
