Use color images as dictionary keys, not bitmaps.

pixelocr/image.py

@@ -164,18 +164,17 @@ class Image(object):
     def isspace(self):
         return not _is_nonblank(self.bitmap)
 
-    def tostring(self):
-        """Serialize the image as a string."""
-        height, width, *_ = self.shape
-        shape = '{}x{}'.format(height, width).encode('latin1')
-        bitmap = np.packbits(self.bitmap).tostring()
-        return shape + b':' + bitmap
+    def serialize(self):
+        """Serialize the image as some hashable object."""
+        bitmap = self.data.astype(np.uint8).tostring()
+        return self.shape, bitmap
 
     @classmethod
-    def fromstring(cls, data):
-        """Deserialize an image from a string."""
-        # TODO
-        raise NotImplementedError
+    def deserialize(cls, obj):
+        """Deserialize an image."""
+        shape, data = obj
+        array = np.fromstring(data, dtype=np.uint8).reshape(shape)
+        return cls(array)
 
     def unframe(self, width=2):
         return Image(self.data[width:-width,width:-width])

pixelocr/page.py

@@ -187,7 +187,7 @@ class Glyph(PageObject):
     @property
     def key(self):
         """Return a dictionary key uniquely representing this glyph."""
-        return self.elevation, self.image.tostring()
+        return self.elevation, self.image.serialize()
 
     def is_body(self):
         """Return True if the glyph is definitely not diacritic."""