Hybrid ant colony optimization model for image retrieval using scale-invariant feature transform local descriptor

Abstract

An organization uses a symbol as its representation in the market for ease of identification and uniqueness. Logos are used to identify and retrieve the materials, even in a complex environment for further analysis. Algorithms based on support vector machine and neural networks provide better results in retrieval of the document from small dataset. But inlarge data sets the existing models lags in their classification performance. This proposed model uses ant colony optimization (ACO) along with the local descriptor scale-invariant feature transform (SIFT), as a hybrid model for retrieving document from dataset. This hybrid model enhances the performance of the retrieval model in terms of increased efficiency, leading to an accuracy of 95.86% with a high output precision of 97.67%.

Introduction

In 20th century digital world, many organizations private and public sectors are interested in implementing their processes via digital models. They all have digital service rooms to reduce paperwork. Since a manual process is required to consolidate papers, these digital rooms are used to reduce this burden by improving efficiency and simplifying administration processes such as organizing, classifying and retrieving documents. An automatic indexing system is essential for this digital service, automatic classification and distribution of documents. The use of textual information for such process is complex, since all the documents are developed using the same format. Therefore, it is essential to identify the documents based on the logo or a symbolic representation in the layout. In some documents, a bar code or quick response (QR) code is used to give the details of the organization. This salient feature creates broad opportunity for content-based document image retrieval in analyzing documents.

The application of logo-based document retrieval is based on common pattern recognition processes, which differ from traditional processes mainly in the extraction portion. The usual recognition process starts with preprocessing, followed by extraction of a meaningful logo or symbol from the scanned document. After the extraction process, classification is performed to detect the logo and compare it with the existing database in a matching process. This includes a critical consideration of the recognition processes for machine-designed logos and manually designed logos. The current research in image processing and data retrieval is reviewed in order to identify the best retrieval model.

In document analysis, the process is described as two tasks, based on the interest of the researcher in the design process. The initial process is used to identify the logo and then a classification is performed, even if the obtained image is not present in the existing database. The second model uses a different process, as it extracts the logo, and then all the related documents are extracted based on the index values. In both processes, indexing plays a vital role and therefore it is considered an important feature for logo extraction. It is necessary to identify the detection procedure for a given document in order to summarize the documents.

In many logo recognition models, it is evident that the identification process is performed through segmentation. The steps involved in detection methodologies using common processes is given in Fig. 1. This figure shows techniques used in document detection based on various approaches, such as the connected component model, window-sliding model, detection based on recognition model and the use of local descriptors. The features for logo recognition is characterized based on the utilization of resources and are mainly classified into two categories, i.e., structural logo recognition and statistical logo recognition. Structural recognition considers the geometric relationship among the primitives and uses a graphical representation technique for recognition of shape and other contextual features. In the statistical recognition approach, enormous quantities of information from various sources are used. The local and global factors are considered based on the features, and these are categorized into local features and global features. The summations of extracted information from each input domain are considered as local features, whereas global features are concerned with the set of pixel values for a particular region of the document. Local features are used in the logo retrieval process due to their characteristics. The invariants, negative features, curvature and centroid point are considered with respect to local features in the detection model. For these descriptors, the total area and vertical gaps are considered in the detection, along with the color and other features. The bag-of-word features and the Fourier coefficients are considered in some cases in the logo detection model. In addition, many models, such as bispectral feature extraction, wavelet-based feature extraction, and two-level contour representation, are used in research studies related to logo detection. Classification is the next process in logo recognition, and this employs various models, such as those using nonparametric and parametric factors.

The organization of the manuscript is as follows, Section 1 forms the introduction, Section 2 deals with related work, Section 3 gives detailed information on ant colony optimization in the document retrieval process, along with the scale-invariant feature transform local descriptor, Section 4 discusses and evaluates the results and Section 5 concludes the study.