Database Reference
In-Depth Information
4.2.2.2
Mobile Visual Search in Academia
In academia, the workshop on mobile visual search has been gathering researchers
and engineers to exchange various ideas in this field [
109
]. Quite a few research
efforts have been put into developing compact and efficient descriptors, which can
be achieved on the mobile end. Chandrasekhar et al. developed a low bit-rate com-
pressed histogram of gradients (CHoG) feature which has a great compressibility
[
111
]. Tsai et al. investigated in an efficient lossy compression to code location
information for mobile-based image retrieval. The performance is also comparable
with its counterpart in lossless compression [
138
].
On the other hand, contextual features such as location information have been
adopted and integrated successfully into mobile-based visual searches. Schroth et al.
utilized GPS information and segmented searching area from a large environment
of city to several overlapping subregions to accelerate the search process with a
better visual result [
134
]. Duan and Gao proposed a side discriminative vocabulary
coding scheme, extending the location information from conventional GPS to indoor
access points as well as surrounding signs such as the shelf tag of a bookstore, scene
context, etc. [
119
].
Additionally, other researchers targeted practical applications and provided
promising solutions. Takacs et al. proposed a loxel-based visual feature to describe
region-related outdoor object features [
136
]. Chen and Tsai proposed methods on
using image processing techniques to find book spines in order to index book
inventories based on bookshelf images [
112
,
137
]. Girod et al. investigated mobile
visual search from a holistic point of view with practical analysis under mobile
device constraints of memory, computation, devices, power and bandwidth [
120
].
An extensive analysis using various feature extraction, indexing and matching tech-
niques is conducted using real mobile-based Stanford Product Search system. They
demonstrated a low-latency interactive visual search with satisfactory performance.
4.2.3
A Framework of Context-Aware Mobile Visual Search
Aforementioned visual search methods and applications on mobile devices have
demonstrated their merits. Alternatively, it is believed that combining visual recog-
nition techniques with personal and local information will provide contextually
relevant recommendations. Hence, this work describes a mobile visual search model
to suggest potential social activities on-the-go.
Three types of user interactions (i.e., tapping, straight line, and circle gestures)
have been investigated to facilitate the expression of the user intent. Then, the
visual query goes through an innovative contextual visual retrieval model using
the state-of-the-art BoW paradigm, to achieve a meaningful connection to database
images and their associated metadata information. Once the user intent expression
is predicted by such visual recognition, associated textual information of retrieved
Search WWH ::
Custom Search