characteristics would be insigli
varic
Sele
whi
organisms in nature have tried to make
the leaf
Die
the leaf
W
S.No. Name of the Length
plant (the of the
leaf from leaf
which it is
D
w
b b
taken)
Observe them carefully and fill
Width of Colour of Shape/Size Margin of
of the leaf the leaf
people who have tried to study diverse
Answers
Answer:
can you attach the picture
Answer:
: )
Explanation:
Given a query image, the system first uses an interactive
segmentation step to segment the plant from the background.
While we assume that the query pictures will be close
snapshots of the plant, the pictures cannot be assumed to be
free of background (e.g. pot, table, background wall). The
segmentation step ensures the extraction of relevant features, by
focusing on the plant region rather than the background. In the
current system, the segmentation is run semi-automatically: for
the database images, the foreground and background regions
are marked manually until a satisfactory segmentation result
is obtained. In the actual use scenario, the user would have
to mark a few regions to identify some background and
foreground regions; but once those regions are selected, further
operations including segmentation are performed by the system
automatically without needing any additional information.
Using segmented plant image rather than the original image
brings an important gain in the system performance.
The segmented region is used for feature extraction which
consists of well-known color, shape and texture features,
along with some modifications to existing texture matching
techniques and introducing some new shape features. We
studied the suitability of these features, both individually and
in combination, in order to find the best combination for the
given problem. The feature extraction and matching steps are
explained in the following subsections.
3.1. Image segmentation
We use the max-flow min-cut (MFMC) graph cut method
[34] to segment the plant images from the background. The
MFMC technique has recently become one of the most popular
segmentation approaches in computer vision; it efficiently
minimizes an energy defined on a graph constructed over the
image, based on the image descriptors. In the basic graphcut approach, an image is represented as a graph where the
graph nodes are the pixels and the graph edges are formed
between the neighboring pixels in the image. The algorithm
requires seed foreground and background pixels (source and
sink, respectively) to be specified. It then splits the graph into
two disjoint sets S (source) and T (sink), minimizing a cost
functional. The selected seeds form the initial values of the
sets S and T . The assignment of the image pixels into two
disjoint sets corresponds to a binary labeling of the image with
foreground and background regions. The functional is based on
two values: (i) a spatial smoothness term which measures the
cost of assigning the same label (e.g. foreground or background)
to adjacent pixels and (ii) an observed data term that measures
the cost of assigning a label to each pixel.
Once an energy or cost functional is defined as described
above, one can resort to efficient optimization methods that
exist in the algorithms literature. For solving MFMC problem
on directed weighted graphs, the augmenting path algorithm
of Ford and Fulkerson [35], the push-relabel method and a
modified version of the augmenting path method by Boykov
and Kolmogorov [34] were introduced. In our work, Boykov
and Kolmogorov’s technique is utilized since it is efficient and
widely used. There is no parameter expected from the user for
the segmentation; both the query and the database images are
segmented using the same settings. The most important variable
for the graph-cut algorithm is the variance estimator (σ 2) that is
used to calculate the capacity between adjacent pixels (of edges)
in order to maximize the flow. We used the default value for
this variable that is determined according to the image intensity
values. Specifically, σ2 is set according to the variance of the
foreground seed region points and the maximum intensity of
the image.
Currently, the seed and background selection is carried
out manually, using a MATLAB GUI program that we have
implemented. For selecting the seed points more efficiently, the
user selects seed regions by indicating the vertices of polygonal
regions. Defining five polygonal seed regions requires the
selection of 15 points, on average. Figure 2 shows sample
segmentation results on two plant images from our database.
While the segmentation step of the current system is relatively
easy, we plan on adopting a more user friendly interactive
segmentation technique, such as the GrabCut algorithm [36],
in the future.
The Computer Journal, 2010