Down syndrome is the most easily identified and most common genetic disorder. Down syndrome variation of trisomy 21 has characteristics that are common to people with Down syndrome. But, children with this Down syndrome variation have slight differences in the mouth and appearance because they tend to have similar oral features with their parents and siblings so that distinguishing them is quite complex. Therefore, it is important to know in more details the special characteristics of children with Down syndrome. This study aims to create an innovative image processing-based system so that it is more practical to classify people with down syndrome and normal people. The method that can be used is the Local Binary Pattern (LBP) method. This study used 2400 mouth frames of children with Down syndrome and normal children for data training. Then, it also uses 3600 mouth frames for children with Down syndrome and normal children for testing data. The results obtained are the threshold value which gives a good classification of 0.1-0.2 for the three variations of the distance calculation method. The Euclidean and Chebychev methods have an accuracy quality of 100% while the City block method at the threshold of 0.1-0.2 has an accuracy of 91.6. So, it can be said that the most accurate method in this research is the Chebychev, Euclidean, then City Block method.