Database Reference
In-Depth Information
Fig. 4.17
Standard OCR failed to recognize multiple lines of skewed characters, but is successful
after using the “O
+
TILT alignment” procedure. (
a
) original image with O. (
b
) OCR fails without
O. (
c
)OCRwith“O
+
alignment”
based on the PCA result, and then, further aligned by the TILT algorithm before
the OCR process. Figure
4.17
illustrates a successful OCR detection.
Two visual examples are demonstrated in Fig.
4.18
with the visual queries
associated location metadata of (a) Bleecker Street Pizza, located at 69 7th Ave S.
New York. (b) Beef Marrow and Marmalade, located at 97 Sullivan St. New York.
4.4.3
Subjective Evaluation
It is conducted a subjective evaluation on user experience with the
TapTell
system.
A total of 13 people participated the survey, nine male and four female. Eight out of
the total participants had heard of the term content-based image retrieval, and six of
them had heard of a natural user interface. During the survey, they were asked about
the usefulness of and satisfaction with the system based on their experience using
the prototype. The survey scale is ranked from 1 to 5 for usefulness and satisfaction,
where 1 is the least and 5 is the most. Table
4.3
summarizes the survey result.
Question 1 and 2 are about the usefulness of the “O” gesture compared to
segmentation and line-based gestures, and the satisfaction of the “O” interface.
Question 3 and 4 are about visual search satisfaction on duplication/near-
duplication results, as well as semantic similar results. The rate is higher for
the former, which is a fair reflection of the algorithm we took. This is because
we use salient-based SIFT points, which are more suitable for duplication/near-
duplication detection than object recognition.
Search WWH ::
Custom Search