SEARCH PREPROCESSING

 

The search preprocessing is to find the original location of the eye’s features from a face shadow. As an important process of the eye feature extraction and description, the search preprocessing is always time consuming [9-10]. Many experts have used a lot of principles to complete their search. For example, edge projection analysis is typical shadow processing technology. The projection of laternal and vertical image coming from analyzing the face shadows can produce a Histogram chart. Then using the Histogram analysis plus the places of the five sense organs, it is easy to find out the original location of the eyes. Since the eyeballs are darker than other parts of the face, the blob detection based some mathematical exercises on the darkness and brightness contrasts to get the location of the eyeballs.

 

In order to save much search time in this experiment we use checker search. With the checker search, we find the best feature subset and features that are needed before the checker search. Our experiment is to put the CCD camera in front of the faces and make the light shine up. Through many experiments and tests, we find 20 pixel is the biggest diameter of the black round spots in the eyeballs as we make the face shadows binary. We thus take 20×20 checkers to search greatly in the binary shadows, which are composed of many 20×20 checkers. The research steps are from left to right and from up to down. Several features may occur while every 20×20 checker is searching (as Fig.2 shows):

 

Fig. 2 Several features may occur while

every checker is searching.

Feature 1: all dark colors. It may be the colors of the background and the hair.

 

Feature 2: all bright colors. It may be the colors of other parts of faces.

 

Feature 3: the central part are colors while the surroundings are bright.

Feature 4 (a),(b): the darts gather on the right and down side.

 

Feature 5 (a),(b): the darts gather on the right and up side.

Feature 6 (a),(b): the darts gather on the left and down side.

Feature 7 (a),(b): the darts gather on the left and up side.

 

Feature 8 (a),(b): the darts gather on the right side.

 

Feature 9 (a),(b): the darts gather on the left side.

 

Feature 10 (a),(b): the darts gather on the down side.

 

Feature 11 (a),(b): the darts gather on the up side.

 

There may however be several cases where the round black spots of the eyeballs appear in the 20×20 checker (as Fig.3 shows):

 

Fig. 3 There may be several cases where the black round spots of the eyeballs appear in the checkers.

 

It is not necessary for us to search in every checker because we just need to do a detailed template comparison to some important features among the four cases mentioned above. When coming across other features, we can quickly skip over them without detailed template comparison.

 

Sine our search steps are from left to right and up to down, we just do detailed template comparison to 4 most possible feature and skip other features. The method we use is called diagonal-box checker search.

 

First we create two search checkers (as Fig.4 shows):

 

Fig. 4 Two search checkers

 

 

If all gray levels on the line a are smaller than line a’s thresholds, then a=1; otherwise, a=0. The same to line b, line c, line d, and line e. If all gray levels on the line f are bigger than line f’s thresholds, f=1; otherwise, f=0. The same to line g, line h, line i, line j, line k, line l, line m, line n, and line o. The method of diagonal-box checker search is to check the gray level of these lines for searching the locations of the eyes. The detailed algorithm of diagonal-box checker will be discussed in appendix A.

 

 

index