17-03-2014, 06:44 PM
Abstract: This paper proposes a novel method to
identify data patterns that mutually occlude each
other. In pattern recognition problems, occlusion
is a challenge which severely limits the accuracy
of recognition. The technique makes use of an
evolutionary algorithm at a micro level by
extracting a population from the occluded text
pattern and by doing the operations such as
selection, crossover, mutation and reinsertion on
the members in the population. The hidden text
is found out after certain number of generations.
The technique works well even if the occlusion
is mutual, that is, when both text patterns
occlude each other and even if up to 90% of the
pattern is hidden.
Keywords:Occluding text patterns; Similarity
index; Rank of match; Fitness value; Vocabulary
1. Introduction: Occlusion refers to
overlapping of one object by the other.
Though lot of work is being carried out
by researchers to resolve the problem of
occluding images [1,2], the problem of
resolving occluding text has been
relatively less addressed. In this paper
occlusion of text patterns is taken as the
problem and a novel method has been
proposed to resolve the partially hidden
text patterns. The term ‘text pattern ’ in
the context of this paper, refers to a
combination of alphanumeric characters.
The set of all possible text patterns is
referred to as a vocabulary, the sample
of which is shown in figure 1. The aim
of this work is to resolve the patterns in
the vocabulary that mutually occlude
each other to produce a resultant pattern.
In the process of mutual occlusion of
patterns, the resultant pattern may be
very difficult to identify since the
occlusion can be random in nature.
Figure 1: Vocabulary of Text Patterns
Figure 2 shows the random nature of
occlusion extending from simple single
point occlusion to multiple point
complex occlusion. Figure 3 shows
some typical cases in which the patterns
occlude each other to generate a
resultant pattern. The characters that
have been stricken through indicate that
they are occluded by the other pattern.
identify data patterns that mutually occlude each
other. In pattern recognition problems, occlusion
is a challenge which severely limits the accuracy
of recognition. The technique makes use of an
evolutionary algorithm at a micro level by
extracting a population from the occluded text
pattern and by doing the operations such as
selection, crossover, mutation and reinsertion on
the members in the population. The hidden text
is found out after certain number of generations.
The technique works well even if the occlusion
is mutual, that is, when both text patterns
occlude each other and even if up to 90% of the
pattern is hidden.
Keywords:Occluding text patterns; Similarity
index; Rank of match; Fitness value; Vocabulary
1. Introduction: Occlusion refers to
overlapping of one object by the other.
Though lot of work is being carried out
by researchers to resolve the problem of
occluding images [1,2], the problem of
resolving occluding text has been
relatively less addressed. In this paper
occlusion of text patterns is taken as the
problem and a novel method has been
proposed to resolve the partially hidden
text patterns. The term ‘text pattern ’ in
the context of this paper, refers to a
combination of alphanumeric characters.
The set of all possible text patterns is
referred to as a vocabulary, the sample
of which is shown in figure 1. The aim
of this work is to resolve the patterns in
the vocabulary that mutually occlude
each other to produce a resultant pattern.
In the process of mutual occlusion of
patterns, the resultant pattern may be
very difficult to identify since the
occlusion can be random in nature.
Figure 1: Vocabulary of Text Patterns
Figure 2 shows the random nature of
occlusion extending from simple single
point occlusion to multiple point
complex occlusion. Figure 3 shows
some typical cases in which the patterns
occlude each other to generate a
resultant pattern. The characters that
have been stricken through indicate that
they are occluded by the other pattern.