Information Technology Reference
In-Depth Information
11.2 Visual Information Description
To index and search visual information, the images and videos need to be
characterized using quantified visual features. There are a number of visual
features proposed in the literature [5,11]. In this section, we will focus on the
visual features standardized by the MPEG-7 [1,2].
MPEG-7, formally known as Multimedia Content Description Interface,
provides interoperability among systems and applications in audiovi-
sual content descriptions that help to identify, index, and retrieve audiovi-
sual information. See Section 11.3 for the visual information description of
MPEG-7 Visual. It lists a set of visual descriptors for specific visual features
such as color, texture, shape, and motion. The main descriptors are briefly
explained in the following. More comprehensive specification can be found
in ISO/IEC information technology [1,2].
11.2.1 Color Feature Descriptors
One of the most important visual perceptions for humans is color. Color
has proved to be one of the most effective and efficient features for simi-
larity indexing and retrieval and therefore plays a critical role in visual
similarity of content. Several feature descriptors based on color are specified
in MPEG-7.
11.2.1.1 Dominant Color Descriptor (DCD)
This descriptor specifies a set of dominant colors. The dominant color fea-
ture is extracted from a set of pixel color values in the RGB color space and
quantizes the color vectors in the image based on the Generalized Lloyd
Algorithm (GLA). The color vectors are the centroids of color clusters in the
CIE LUV color space, called ColorBin and the number of bins is denoted as
Bin-Num. The dominant colors are then represented in the RGB color space,
which is converted back from color centroids in the CIR LUV color space.
Along with the color vectors, percentage and color variance are specified for
pixels corresponding to the dominant colors. In addition, spatial coherency
is specified in the descriptor as a single value computed by the weighted
sum of predominant color spatial coherency, proportional to the number of
pixels corresponding to each dominant color. The predominant color spatial
coherency describes pixels corresponding to the dominant color that appear
to be a solid color. The extraction algorithm is described in detail in ISO/IEC
information technology [2].
The dominant color descriptor (DCD), as the name suggests, captures the
dominant colors in pictures.
