This article presents a new embeddable method for polyp detection in endoscopic video images and wireless capsule endoscopic images based on physician approach. This approach is twofold: the first step consists in a geometric approach to characterize the polyp's geometric features as size or shape. The second, a texture approach based on a learning process of texture features of polyps using the use of the co-occurrence matrix and their related statistics. For classification, we propose the boosting method which allows us to generate a strong classifier and also the possibility of permanent learning. The performance of the learning approach over a database of 300 images, is characterized by a sensibility of 90,6%, a specificity of 95,0% and a false detection rate of 4,8%. All the different steps of the algorithm were meticulously chosen to facilitate the future hardware implementation of the treatment on the capsule for wireless endoscopy.