Article Preview
Top1. Introduction
The automatic reading systems of documents are widely used to store, sort and search any important information from paper-based documents. The objective of document image analysis is to extract and recognize the textual information (using OCR system) and graphics components in the document. A typical OCR system consists of the following steps:
- •
Step 1: Image preprocessing (page's restoration, quality improvement, noise reduction, contrast enhancement, deskewing of the document, page magnification).
- •
Step 2: Foreground/Background separation (Binarization) using global or local methods.
- •
Step 3: Component separation and segmentation, using a pyramidal analysis (extraction of the physical and logical layout of the document, text block detection, printed or handwritten recognition)
- •
Step 4: Optical recognition of text.
As described in (Gaceb et al., 2008), in general there does not exist an ideal and generic segmentation because of the difficulty of parameter choice on document images of very diverse quality. There are always several possible segmentations. We can show that the problem of segmentation is usually a poorly formulated problem. A good segmentation mechanism will be able to provide a good recognition on a wide variety of documents. Therefore, it should be able to simplify the image content without degrading it, by avoiding precipitated and irreversible choices. We note that the future of image preprocessing and segmentation is in the downstream control by using evolutionary approaches. It is in this context that the PSO algorithm was introduced to facilitate, accelerate and optimize the choice of preprocessing settings applied to images of documents. The evolution of a global optimum search is guided by the quality of binarization. Image foreground and background separation is also an essential step in a variety of image analysis, OCR and computer vision tasks (Gaceb et al., 2008).
In our study, we have chosen PSO algorithm for the following reasons; it comprises very simple concepts, and the paradigm can be implemented in a few lines of computer code. It requires only primitive mathematical operators and it is computationally inexpensive in terms of both memory requirements and execution time.
The rest of this paper is organized as follows: In section 2, we give a general overview on more methods presented in the literature, which are related to our problem, especially the preprocessing of degraded images. In section 3, we present the PSO paradigm principle, the equations and the algorithm. In section 4, we present the theoretical principle of the Mean Shift method. The section 5 describes our new approach and the contributions in terms of preprocessing of document images. This last section is organized into two parts; the first one describes an adaptive mechanism of Mean Shift (AMS) method for smoothing and filtering and the second one explains the role of PSO algorithm in the image filtering and smoothing by using an improved Mean Shift method. At the end of this paper (section 6), we present and discuss some experimental results obtained from applying our propositions on a set of images of the degraded documents from DIBCO11 Dataset before concluding the paper in section 7.