Face recognition has been one of the most interesting technology to study for many researchers. It allows a huge number of face images to be recognized in just a short amount of time, rather than recognizing each image individually through a normal human’s eyes. Its idea is generally based on the assumption that each individual has a unique identity that can be distinguished from one individual to another. However, in the real world, there will be individuals who have faces similar to each other. They are referred as "look-alike" faces.

This research is conducted to recognize look alike faces. By doing so, real-world applications such as searching for missing person, criminals or fugitives can become easier. By recognizing look alike faces, the running process will save much more time because unqualified faces can be removed as they are not needed to be used for further processing. Because the data on the actual problem is usually very large, the reduction of such data can reduce the cost. A database for look-alike faces has already been made by Lamba et al. in 2011, this will be the database used in this research.

The authors will develop face recognition algorithms using and Support Vector Machine. This method has been used and resulted in high accuracy for face recognition by other researchers. It is believed that not all the features of a face image are necessary for the algorithm to do its job. Reducing the number of features will also reduce the dimension which will save the space and cost needed. Therefore, the authors will compare the algorithms with feature selection and the algorithm with no feature selection. The feature selection method that will be used in this research is Chi-Square Feature Selection.

Support Vector Machine is one of the most popular machine learning algorithms, while Chi-Square has been used for face recognition to reduce dimension for over years. The use of classifier without feature selection may result in lower accuracy. Therefore, by combining the algorithms above, it is believed that the accuracy will be higher than the previous researches conducted for the case of identifying look-alike. The authors of this research are still developing the algorithms and expect the accuracy to reach 80%. Face images which were correctly classified and incorrectly classified by the algorithms will also be shown. By knowing the failure of the classification, the authors hope to conduct further researches to develop more accurate algorithms for face recognition to identify look-alike faces.