2003 Volume 32 Issue 4 Pages 369-377
A method for fast picture search is proposed. Additive noise and quantization noise are statistically analyzed. Sample points can be qualified by use of the knowledge about those noises and randomly selected as pixels of low contrast. A proposed test quantity is proved to have an inequality to the χ 2 statistic calculated from the sampled brightness differences defined by a template picture and an object from the scene. Search evaluation is performed by increment sign correlation which is known as of fast computation. A modified version using a parameter is presented in the case of large noise. The feasibility of the proposed method is shown by experimental results.