The motivation of our study is to provide algorithmic appro-aches to distinguish proper patterns, from garbage and erroneous patterns in a pattern recognition problem. The design assumption is to provide methods based on proper patterns only. In this way the approach that we propose is truly versatile and it can be adapted to any pattern recognition problem in an uncertain environment, where garbage patterns may appear. The proposed attempt to recognition with rejection combines known classifiers with geometric methods used for separating native patterns from foreign ones. Empirical verification has been conducted on datasets of handwritten digits classification (native patterns) and handwritten letters of Latin alphabet (foreign patterns).