The handwritten character recognition in digital images is an important and challenging area of study in Computer Vision, with several possibilities for applications to facilitate the daily work of the people. This paper presents an algorithm for handwritten character recognition with two proposed approaches. The first proposal complements earlier work by some of the authors of this article, including 290 new attributes, based on histograms, Zoning and transformed Hit-or-Miss. The second proposal uses 79 attributes, obtained from frequency information, distance-edge character and densities, which performs classification using an approach based on maximum and minimum values of each attribute for each character type, and a neural network Multilayer Perceptron. The large number of attributes contributes to a more precise discrimination of characters, on the other hand, the extraction of these descriptors is easy because only performs the pixels counting. Thus, the processing time in this task is reduced. Although the classification using a Multilayer Perceptron neural network achieved a higher hit rate, the processing time of the maximum and minimum limits based classification is smaller, allowing its use in applications where the processing time is critical.