Handwritten Word Spotting in Old Manuscript Images using Shape ...

More documents

Recommendations

Info

Figure 5: We present two approaches based in word spotting. Both have the same firsts steps. first to quickly reject an important number of non similar words (first level) and do the intensive search with more discriminant features (BSM) in the second level with a reduced number of target words. The second approach is oriented to pseudo-structural features. The descriptor used in this approach is characteristic Loci feature and the indexation structure is constructed using a table, where each column is each observation of the documents, and the rows are the features of the words. Each word, or character, is composed by several features, and it is not significant where they appear inside the image. This approach uses features based in Loci Characteristics [3; 4; 8]. Given a word image, a feature vector based on Loci characteristics is computed at some characteristicpoints. Some approaches of the literature have used the background pixels of the image. Other approaches have used the foreground pixels, and even some approaches have used the contour or the skeleton of the images. Loci characteristics encode the frequency of intersection counts for a given characteristic-point in different direction paths starting from this point. Loci vectors extracted from the words of the image database are stored in a hashing structure. Afterwards, the word spotting is performed by a voting process after Loci vectors from the query word are indexed in the hashing table. Let us describe the different steps of the two developed approaches. Both approaches have the same preliminary steps. They consist in a pre-processing step, where the documents are segmented and extracted the words of them, in a fast rejection, where bad words are discarded, and noise removal, where the noise of the image is removed and the bounding box is fixed to the contour of the image. These preliminary steps are explained in the section 6. Section 7 explains the first approach developed and the section 8 the second one. 10
6. Preliminary steps 6.1 Pre-processing Modelling the human cognitive process to obtain a similar computational methodology for handwritten word segmentation is quite difficult due to the following characteristics. The handwriting style is usually in cursive or discrete. In the case of discrete handwriting, characters are joined to form words, but, unlike the machine printed text, handwritten text is not uniformly spaced. The size of the characters along the words of the document is different (this is a scale problem). Ascenders and descenders are regularly connected and words present different orientations. Documents are often degraded due the ageing or other reasons. Another reason is the presence of show-through or bleed-through effects explained above. Some of the main problems of our historical documents are that they have been written by several authors (every two years the writer changes), noisy (stains, shadows, bleed through, etc.), margins, etc. The documents to be used in our experiments present some of the above commented drawbacks, like ascenders and descenders connected, different sizes of character, etc. But a good characteristics of these documents is that they are well structured. As we have commented in section 2, each document has three parts, and the objective is to work with the marriage licenses. The steps of the pre-processing are: binarization of the documents, page segmentation, layout segmentation, segmentation of the lines and, the last step, the word segmentation (Fig. 6). Let us in the following subsections describe the details of these steps. Figure 6: Pre-processing steps. 11
Page 1: MASTER IN COMPUTER VISION AND ARTIF
Page 4 and 5: a step forward towards shortening t
Page 6 and 7: a classification of the training se
Page 8 and 9: (a) 1617: index of volume 69 (b) 17
Page 10 and 11: The models can then be used to retr
Page 14 and 15: 6.1.1 Binarization The binarization
Page 16 and 17: partial Gaussian derivatives along
Page 18 and 19: 7. Pixel-based descriptors organize
Page 20 and 21: layer uses an automatic method. It
Page 22 and 23: Table 1: Intervals for each directi
Page 24 and 25: The segmentation process experiment
Page 26 and 27: Table 2: Pre-processing results. Th
Page 28 and 29: Figure 18: Distribution of the obse
Page 30 and 31: Figure 20: Comparative using differ
Page 32 and 33: Figure 22: Classification process u
Page 34 and 35: use foreground pixels, because the
Page 36 and 37: [15] G. Nagy. Twenty years of docum
Page 38: Figure 27: SOM using characteristic

Handwritten Word Spotting in Old Manuscript Images using Shape ...

Create successful ePaper yourself

Delete template?

Save as template?