Generic filtering and removing artefacts from document images using unsupervised PSO optimisation

The advancements in the field of analysis and optical recognition of document images have accelerated recently due to the many emerging applications which are not only challenging but also computationally more demanding, such as mail and document sorting, automatic classification of documents, handwriting and script recognition, etc. In this paper, our contribution focuses on preprocessing of these applications: smoothing and filtering of degraded document images using a new adaptive mean shift algorithm based on the integral image. The great difficulty of parameter setting of this approach requires solving of complex optimisation problems using metaheuristic algorithms. Our goal is to demonstrate the contribution of the particle swarm optimisation (PSO) method to improve the quality and the parameter setting of the developed preprocessing approach. We tested and compared two types of objective functions (supervised and unsupervised) and demonstrate the effectiveness of the optimisation in an unsupervised context.