The application of robot system by using gray scale images and pre-process using the Wavelet Haar method, often applies some error to calculating similarity matrix values in the identification of a face image. The results for 1000 samples image, the processing time required takes 45-60 second. Thus, it requires the calculating similarity with color, texture, prewitt gradient (call these Gx and Gy). Retrieval (CBIR) is thoughtly more advantage, with its popularity and generating test by using time and accuracy level parameters. Content-Based in Content Based Image Retrieval (CBIR) works by measuring the similarity of the query image with all the images that exist in the data base, so that the query cost is proportional to the number of images in the database. The search for the most similar image has a range search by performing image classification that aims to reduce the query cost in CBIR. Implementation of web-based attendance system is using in calculating the similarity. This study is aimed to enable to extract the color features, texture and the edges of the face image by using prewitt gradient. The results of the feature extraction process are then be used by the software in the learning process and in calculating similarity. The learning image contained in 5 classes of features, images stored in data base query are 1000 bmp and jpg image, and the image of the test sample will be sized of 400 x 400. The results showed that the color feature combination, texture and edge detection with prewitt gradient magnitude, showed a significant effect i.e. higher accuracy level than by using gray scale image with Wavelet Haar Method. The face image Calculating similarity takes longer in processing time, which requires 20-40 seconds in time processing.