Face detection in a compressed domain

We focus on the Viola-Jones face detector in the Discrete Cosine Transform (DCT). In order to avoid typical pitfalls due to features in the DCT domain we propose to merge the integral image stage with the decompression process, thus saving time for both operations. The proposed method saves 128 additions and 224 multiplications on an 8×8 block for a simple integral image. We propose a similar method for fast feature contrast adjustment by computing the squared integral image using DCT and simple integral image coefficients. These methods are faster and more transmission error resilient than classical methods.