A new methodology for evaluation and benchmarking of skin detector based on ai model using multi criteria analysis (IR)
This study aims to develop a new multi-criteria decision analysis methodology for skin detector evaluation and benchmarking based on artificial intelligence models. Two experiments were conducted. The first experiment comprised two stages: (1) Adaptation of the best previous case of skin detection a...
Saved in:
| Main Author: | |
|---|---|
| Format: | thesis |
| Language: | eng |
| Published: |
2018
|
| Subjects: | |
| Online Access: | https://ir.upsi.edu.my/detailsg.php?det=4491 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study aims to develop a new multi-criteria decision analysis methodology for skin detector evaluation and benchmarking based on artificial intelligence models. Two experiments were conducted. The first experiment comprised two stages: (1) Adaptation of the best previous case of skin detection approach utilizes multi-agent learning based on different color spaces. This stage aimed to create a decision matrix of various color spaces, and three groups of criteria (i.e., reliability, time complexity, and error rate within dataset) to test, evaluate and benchmark the adapted skin detection approaches. (2) Performance of multiple evaluation criteria for skin detection engines, this stage included two key stages. First, the correlation between criteria to investigate their relationship and determine their degree of correlation. Second, the performance analysis of criteria to identify the factors that affect the behavior of each criterion. The second experiment utilized a new multi-criteria decision-making by adopting the integration of TOPSIS and AHP to benchmark the results of skin detection approaches. In the validation process, multi-criteria measurement was used to calculate the tradeoff for different criteria. Color spaces assessment were conducted to determine the best color spaces with adaptive skin detection engines. Moreover, mean and standard deviation values for thresholds were calculated to select the best color space. Two groups of findings were provided. First, the overall comparison of external and internal aggregation values in selecting the best color space, that is the norm RGB at the sixth threshold. Second, (1) the process proves that the distribution of color spaces with its threshold values affects the behavior of the criteria determined as a trade-off between the criteria according to their weight distribution. (2) The YIQ color space obtains the lowest value and is the worst case, whereas the norm RGB color space receives the highest value and is the most recommended. (3) The best result achieved at the threshold = 0.9. Thus, the implications of this study benefit individuals, research centers, and organizations interested in skin detection applications. Moreover, it provides benefits to software developers working in industrial companies and institutions in developing different techniques and algorithms with different applications. |
|---|