Automated document recognition systems are critically important for remote identification; however, their vulnerability to software-based manipulations of document images necessitates the development of robust methods for detecting such forgeries. This work addresses the task of Image Manipulation Detection and Localization (IMDL), for which there remains no universally accepted formal problem statement, hindering progress evaluation and reducing research consistency. In this study, we propose a formalization of the IMDL task through image provenance verification, where the provenance is defined as a sequence of operations that form the image, from its capture to potential modification. The proposed formal problem statement is particularly relevant for document recognition systems, where a priori assumptions exist about image sources and editing methods. The results of this work establish a foundation for the systematic development of IMDL algorithms. The method can be applied in modern CAD systems to analyze the content of recognized textual files.