Zulkifli, Musa (2014) Intelligent monitoring system for the classification of customers using the hybrid face detection. Masters thesis, Universiti Kebangsaan Malaysia (Contributors, UNSPECIFIED: UNSPECIFIED).
|
PDF
ZULKIFLI BIN MUSA.PDF Download (990kB) | Preview |
Abstract
Monitoring systems in most outlets in Malaysia only capable of recording activities on the environment and the safety of the premises were handed over to the security guards that have limited capabilities. This is a factor for the presence of masked criminals is often successful in their robbery mission. Therefore, a detection system that exposed human skin needed for the classification of each customer. There are three objectives of the construction of this facial skin detection. The first is the presence of customers. Second, detect the head in normal standing customers. Third, tracking customers exposed skin and classified into 5 groups. Customer presence is detected by Background Subtraction (BS) method. The method requires two images of the reference image and the current image to separate the foreground (FG) pixels of background (BG) pixels. Morphological methods were used to improve the quality of binary image objects. Detection of the customer's head is done by using the Circle Hough Transform (CHT). This method mapping template circle at each pixel boundary objects and produce CHT matrix that containing the coordinates and radius of the 10 best circles. Next, the selection of a circle for the head has done according Y min in the CHT matrix. Detection of the skin is done using a new and efficient method is known as the Skin Gray (GS). It can detect the diversity of human skin tones like bright skin (Kulit Cerah, KC), tanned skin (Kulit Sawo, KS), and dark skin (Kulit Gelap, KG) to display the image contrast between skin and non-skin in grayscale. Image contrast is produced by multiplication coefficient of GS in each image component (RcxR +Gc+GxBcxB). Three coefficients were used GS; coefficient of the Red Component (Red Coefficient, Rc), the coefficient of the green component (Green Coefficient, Gc), the coefficient of the blue component (Blue Coefficient, Bc). The analysis conducted has found the GS coefficients for each group of skin and choose a single GS coefficient suitable for all skin tones. The coefficient of selected single GS is Rc = 1.00, which Ge = 0.50, Bc = 0.50. Separation of skin and non- skin area on the image contrast is done by setting range of the threshold value (low threshold (THR) and high threshold (THT)). The analysis conducted has found a range of values for each group threshold skin and choose a range that is suitable for all skin tones. Single threshold value range is; THR = 18, and THT = 113. Classification customers into 5 groups based on the range of skin pixels percentage (Peratusan Piksel Kulit, PPK) on the customer's head. Range PPK classified to 100-41, 40-31, 30-21, 20-11, and 10-0 which is the range for the KA, KB, KC, KD and KE. The effectiveness of the GS method for detecting the diversity of the skin is done by calculating the average percentage of the set (Purata Peratusan Set, PPS) accounted for 92.8 %. Overall, the ability of exposed skin detection system evaluated using the method sensitivity and specificity of 27 different sets of images. Results showed an average sensitivity reached 89 %, while achieving 97 % average specificity. In conclusion, the combination of methods used successfully to classify customers into groups depending on the vulnerable skin.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Security systems; Image processing |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) |
| Depositing User: | Ms. 'Arifah Nadiah Che Zainol Ariff |
| Date Deposited: | 21 Mar 2016 04:20 |
| Last Modified: | 24 Aug 2021 02:18 |
| URI: | http://umpir.ump.edu.my/id/eprint/12064 |
| Download Statistic: | View Download Statistics |
Actions (login required)
 |
View Item |