
i
i
i
i
i
i
i
i
250 12 Compression for Cloud Photo Storage
After the proposed feature-based prediction, the residue images are further com-
pressed by transform and entropy coding to generate the compressed data. In our
implementation, these two modules come from HEVC. In the following sections,
we will mainly discuss the proposed feature-based prediction structure and feature-
based inter-image prediction in details.
12.4 Feature-Based Prediction Structure
Let us assume a correlated image set as I = {I
1
,I
2
,··· ,I
N
}, where N is the num-
ber of images in the set. For every image I
n
, there is a set of SIFT descriptors as
F
n
= {f
n
(1), f
n
(2),··· , f
n
(K
n
)}. In other ...