Application Programs and Systems - Heterogeneous Multicore Processor Technologies for Embedded Systems - page 213

Table 6.7 Measured average execution time of Linux system proc esses
Process Time consumed (s)

Initialization 0.0506

Still-image capturing 0.0088

Face-region display 0.0381

Face detection 1.6072

Thumbnail display 0.0176

Similar-image display (top ten images) 0.0798

Similar-image database
access

Registering 0.6348

Feature calculation and search 0.5857

Deletion 0.2530

a mouse connected via the USB interface is detected by a device driver embedded
in the Linux kernel. The device driver outputs on/off values of each mouse button
and the distance of mouse movement. The mouse event detection classifies three
events of mouse button operations: PUSH, REPEAT, and RELEASE. Furthermore,
it converts the distance into the axis of coordinates. The internal event generation is
processed in accordance with the values generated by the mouse detection. The
defined internal events include no events, still-image capturing, face detection, sim-
ilar-image display, similar-image search, similar-image registering, and similar-
image deletion.
When a mouse event is detected on a video-plane area, the still-image capturing
event is generated, and a still image captured from decoded video images is
obtained as a still-image object. The graphics plane is updated in order to display
the newly captured image. Then the area range of the image selected by the mouse
is trimmed. The trimmed image is treated as a face-region image object, and the
graphics plane is updated again. The face detection uses the face-region image
object, and a frame shape is drawn on the area of the detected face. When a mouse
event is detected on the still-image object or the similar-image object, the face
detection is carried out by using these two objects. When a mouse event is detected
on the thumbnail image object, a thumbnail image shown in the event is displayed
as a similar image. When one is detected on a framed face of the face image object,
the face-framed part of the image is trimmed. The trimmed image is converted in
the image format in order to calculate the feature value, and the calculation is per-
formed. Then the similar-image search is carried out by using the calculated fea-
ture value, and the top ten similar images are displayed. When a mouse event is
detected on the framed face, the face image is registered on the similar-image data-
base. When one is detected on a thumbnail image, an entry of the image is deleted
from the database.
The execution time of each process on the Linux system was measured. Table 6.7
lists the average time for the processes. The face detection required is 1.6 s, and the
access of the similar-image database took more than 0.5 s. The time for such processes