Database Reference
In-Depth Information
Fig. 4.16
The time analysis of the
TapTell
system as well as the visual search, based on the
restaurants dataset. (
a
) Total time spent of the
TapTell
system. (
b
) Visual search time spent
4.4.2.6
Improved OCR from “O” and Visual Examples
Besides the visual content, Optical Character Recognition (OCR) is another impor-
tant means to help mobile users to understand their visual intents correctly. It plays a
vital role in translating from the visual feature to the text feature. However, most of
the OCR techniques are sensitive to the orientation of visual input. If characters are
skewed in a certain degree, current OCR techniques cannot successfully recognize
the correct characters. However, such a difficulty can be alleviated by using a
transform invariant low-rank textures (TILT) algorithm to align the severely tilted
characters properly [
141
].
It is found that one of the byproducts from the “O” gesture is that it can achieve
better OCR performance, if the estimation results of two principal components are
utilized by the PCA in Sect.
4.3.2
. Once the original text region is selected by the
“O” gesture, those characters are first aligned by performing rotation alignment
Search WWH ::
Custom Search