From b0320e055d8c3cb48230390c3bc0587da1fb9d0e Mon Sep 17 00:00:00 2001
From: Albert Cervera i Areny <albert@nan-tic.com>
Date: Fri, 13 Jun 2008 20:31:14 +0200
Subject: [PATCH] Added our own trigram implementation. Now it's used in
 recognizer. Recognizer usage of trigram still untested but should most
 possibly work.

---
 __init__.py   |  7 ++++-
 recognizer.py |  7 +++--
 trigram.py    | 72 +++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 81 insertions(+), 5 deletions(-)
 create mode 100755 trigram.py

diff --git a/__init__.py b/__init__.py
index 2ead901..082da09 100644
--- a/__init__.py
+++ b/__init__.py
@@ -16,4 +16,9 @@
 #   Free Software Foundation, Inc.,
 #   59 Temple Place - Suite 330, Boston, MA  02111-1307, USA. 
 
-
+from template import *
+from ocr import *
+from recognizer import *
+from scanner import *
+from document import *
+from barcode import *
diff --git a/recognizer.py b/recognizer.py
index 5bd2bf1..6076525 100644
--- a/recognizer.py
+++ b/recognizer.py
@@ -23,6 +23,7 @@ from ocr import *
 
 from template import *
 from document import *
+from trigram import *
 
 import tempfile
 
@@ -130,7 +131,7 @@ class Recognizer(QObject):
 
 	# Tries to find out the best template in 'templates' for the image file given by 'image'
 	# TODO: Should be reestructured to use function scanWithTemplate()
-	def findBestTemplate( self, cr, file, templates ):
+	def findBestTemplate( self, file, templates ):
 		self.scan( file ) 
 		max = 0
 		bestDocument = Document()
@@ -157,9 +158,7 @@ class Recognizer(QObject):
 					print "Jumping %s due to type %s " % ( templateBox.name, templateBox.type )
 					continue
 				matcherBoxes += 1
-				#cr.execute( 'SELECT similarity(%s,%s)', (translate(text), translate(templateBox.text)) )
-				cr.execute( 'SELECT similarity(%s,%s)', (text, templateBox.text) )
-				similarity = cr.fetchone()[0]
+				similarity = Trigram.trigram( text, templateBox.text )
 				score += similarity
 			score = score / matcherBoxes
 			if score > max:
diff --git a/trigram.py b/trigram.py
new file mode 100755
index 0000000..456da9a
--- /dev/null
+++ b/trigram.py
@@ -0,0 +1,72 @@
+
+class Trigram:
+
+	# Returns a list of the trigrams of a sentence. That is, the list of
+	# all trigrams of each of the words in a string. Words are currently 
+	# splitted by the space character only.
+	@staticmethod
+	def trigramList( text ):
+		words = text.split( ' ' )
+		l = []
+		for x in words:
+			for y in Trigram.wordTrigramList( x ):
+				l.append( y )
+		return l
+
+	# Calculates the list of trigrams contained in a word. If you feed
+	# this function with an string with spaces they'll be treated like
+	# normal characters. The usual trigram function is trigramList() which
+	# returns trigrams for all of it's words.
+	@staticmethod
+	def wordTrigramList( text ):
+		l = []
+		size = len(text) + 1
+		text = '  ' + text + ' '
+		for x in range(size):
+			l.append( text[x:x+3] )
+		return l
+
+	# Calculates similarity between two strings using a trigram algorithm.
+	# This is based in PostgreSQL pg_trgm implementation though in some cases
+	# you'll get different results. For example trigram( 'abcabc', 'abc' ) 
+	# returns 0.3 here and 0.67 in PostgreSQL's version.
+	# There's also a commented alternative for the final calculation of the
+	# distance.
+	@staticmethod
+	def trigram( text1, text2 ):
+		l1 = Trigram.trigramList( text1.lower() )
+		l2 = Trigram.trigramList( text2.lower() )
+		size1 = len(l1)
+		size2 = len(l2)
+		p1 = 0
+		p2 = 0
+		count = 0
+		while p1 < size1 and p2 < size2:
+			if l1[p1] < l2[p2]:
+				p1 += 1
+			elif l1[p1] > l2[p2]:
+				p2 += 1
+			else:
+				p1 += 1
+				p2 += 1
+				count += 1
+
+		return float(count) / float( size1 + size2 - count )
+
+		# Here another way of calculating the similarity
+		#if size1 > size2:
+			#return float(count) / float( size1 )
+		#else:
+			#return float(count) / float( size2 )
+				
+	
+if __name__ == '__main__':
+	print Trigram.trigramList( 'abc' )
+	print Trigram.trigramList( 'hola' )
+	print Trigram.trigramList( 'adeu manalet' )
+
+	print Trigram.trigram( 'abc', 'abc' )
+	print Trigram.trigram( 'abcabc', 'abc' )
+	print Trigram.trigram( 'abcdef', 'abc' )
+	print Trigram.trigram( 'abcdef', 'bcd' )
+	print Trigram.trigram( 'bcdef', 'abc' )