In a face image containing objects sometimes face has a color similar to the background color or objects that are nearby. This causes the system to detect any objects in the face in an image. This study wants to try to overcome these problems. The approach used in this study is a dynamic image segmentation. The segmentation will produce region-region are then used as input for the neural network. From the experiments conducted, the method used is good enough to detect faces. The results showed that the approach used in this study can detect all of the data that had trained, while for the data that has not been trained detection rate reached 70%.