Unsupervised Classification of Structurally Similar Document Images

Abstract

In this paper, we present a learning based approach for computing structural similarities among document images for unsupervised exploration in large document collections. The approach is based on multiple levels of content and structure. At a local level, a bag-of-visual words based on SURF features provides an effective way of computing content similarity. The document is then recursively partitioned and a histogram of codewords is computed for each partition. Structural similarity is computed using a random forest classifier trained with these histogram features. We experiment with three diverse datasets of document images varying in size, degree of structural similarity, and types of document images. Our results demonstrate that the proposed approach provides an effective general framework for grouping structurally similar document images. I. INTRODUCTIONAssociating images with similar structural characteristics is a first and important step in mining visual information from large collections of images. Having an effective image similarity measure makes searching and browsing large repositories of images easier as clustered images can be collectively tagged with attributes and indexed for search and retrieval tasks. Clustering search results for queries on document image collections, or performing near-duplicate detection are examples of applications where good similarity measures are needed.

Evaluation of deep convolutional nets for document image classification and retrieval

“…The first version of the dataset, listed in the results as SmallTobacco, is a sample of 3482 images from the collection, selected and labelled in another work [20]. This version of the dataset was used in a number of related papers [20,22,17]. Each image has one of ten labels.…”

“…Each dataset was split into three subsets for the purposes of experimentation. The SmallTobacco dataset was split as in the related work [20,22,17]: 800 images were used for training, 200 for validation, and the remainder for testing. Since this is a small dataset, 10 random splits in those proportions were created; results reflect the median performance from those splits.…”

Identity Documents Classification as an Image Classification Problem

“…Local feature points, such as SIFT [21], are widely used as local features due to their description capabilities. Regarding the second step, image encoding, BOW were originally used to encode the feature point's distribution in a global image representation [12,16]. Fisher vectors and VLAD later showed improvement over the BOW [23,14].…”

“…the structure of text blocks, figures, and tables. Such descriptions are finally used to perform classification [2,16] or to compute similarities [8,24]. The second type 1 , Ahmad Montaser Awal 2 , and Teddy Furon 1…”

Fine-grained classification of identity document types with only one example

“…In our application, however, the robust detection of text zones is a difficult task in itself for the same reason the OCR fails. Kumar and Doermann [11] use a bag-of-words approach with SURF features combined in hierarchical histograms in order to visually compare documents. While they focus on black and white documents, we also incorporate color documents.…”