One challenge in text processing is the treatment of case insensitive documents. The traditional approach is to re-train a language model excluding case-related features. This paper presents a preprocessing case restoration approach, based on HMM trained on a large raw corpus of case sensitive documents. Benchmarking shows: this approach (i) outperforms the feature exclusion approach for NE, (ii) leads to very limited degradation for parsing and relationship extraction. Additional advantages include: (i) reduced system complexity since each text processing module does not need two models; (ii) wide applicability: the restored text can feed both statistical model and rule based system.