| Budget Amount *help |
¥3,000,000 (Direct Cost: ¥3,000,000)
Fiscal Year 1999: ¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 1998: ¥1,700,000 (Direct Cost: ¥1,700,000)
|
|---|
| Research Abstract |
A method to realize robust face recognition of a face in a scene under different light direction is investigated, then, the following results are introduced.
1. First, a special rotating table is made, rotation angle of which can be controlled by a computer. From a subject on the table, face images with different light direction of every 5 degrees over the all angle are captured. Maximum direction is 45 degrees upward/downward and left/right.
2. Face area is represented by 12 x 12 monochrome mosaic, Fisher's discriminant analysis is then applied to the face images represented by the mosaics.
3. Face images of light directions of every 15 degrees are used as dictionary, unknown face images of every 5 degrees are then tested. High recognition rate 98.1% for 5 persons is obtained.
4. For the reduction of number of dictionary data, face data of middle angle of light direction are deleted. However, average recognition rate drops to 63.3%. Then, by interpolating deleted part from both sides angle data, average recongition rate rises to 77.1%.
5. Moreover, mosaic blocks, intensity of which considerably depend on change of light direction, are omitted from the face area to be computed. Additionally, face data of middle angle of light direction are deleted but interpolated from both sides angle data. As a result, average recognition rate of 94.6% is achieved for changes of 45 degrees left and right and up and down.
|
