In this paper, we propose a new character extraction algorithm for local and parallel processing. This algorithm works under a severe constraint: each pixel in a processed image must be derived using only information of its neighboring pixels. This constraint is very important for a low cost device such as a mobile camera, because it makes possible to process each pixel in parallel. The proposed algorithm consists of a transform process, a base image creation process, a noise extraction process, a character candidate extraction process, and a character extraction process. In the transform process, RGB values of the input images are transformed to brightness value. In the base image creation process, images are smoothed with preserving edges. After the smoothing, images are binarized based on edge information. A Laplacian filter is used for edge detection. The proposed algorithm does not require a brightness threshold to binarize images. In the noise extraction process, binarizing is performed after character elimination. Maximum filter and minimum filter are used for character elimination. In the character candidate extraction process, character candidates are extracted from output images of the base image creation process and the noise extraction process. In the character extraction process, characters are extracted based on the size of each region. In this process, pixel values of each region are moved and added toward center of each region. Based on the size of each region, proper regions for characters are extracted. The proposed algorithm is represented by local and parallel image processing and has been tested using 100 scenery images. The followings have been confirmed from the results of the computer simulations: 1)The proposed algorithm can extract characters from scenery images better than the conventional algorithm: 2)The algorithm can be specialized for arbitrary size characters.