07-06-2012, 04:34 PM
Leaf Extraction from Complicated Background
Leaf Extraction from Complicated Background.pdf (Size: 784.85 KB / Downloads: 0)
INTRODUCTION
Plant species classification and identification is one of the
difficult and important tasks in agriculture due to its variety
and different field conditions. The traditional method based on
expert’s manual labeling is very time consuming. With the
development of computer science and information technology,
machine vision is a promising technology to solve some tasks
in the agriculture such as species identification, plant
modeling and herbicide application for weed control [1, 2].
The framework based on this technology uses the vision
sensors such as CCD to capture the plant images firstly. The
digital image processing and pattern recognition techniques
are then applied for feature extraction and image analysis.
THE PROPOSED LEAF EXTRACTION ALGORITHM
For the leaf images with complicated background, the
traditional methods such as thresholding, edge detector and
morphological processing cannot perform well to segment the
target leaf from the complicated background due to their
limitations. We propose an automatic leaf extraction algorithm
by using the marker-controlled watershed segmentation three
times and the solidity measure. Instead of using the RGB color
image, our algorithm is performed on the HSI color space
which consists of Hue, Saturation and Intensity values.
Marker-controlled watershed segmentation
The watershed method [8] is an image segmentation
method that divides an image into some regions based on the
topology of image. A grey-scale image can be considered as a
topographic surface. If the surface is flooded from its minima
while avoiding the merging of the waters coming from
different sources, the image is partitioned into two different
sets: the catchment basins and the watershed lines.
CONCLUSION
This paper has proposed a new algorithm which can
automatically extract the target leaf from the color images with
complicated background. We apply the marker-controlled
watershed segmentation to the Hue, Intensity and Saturation of
HSI color images separately which contain both the intensity
and color information. The target leaf is then extracted from
each image segmentation. Finally, the solidity measure is used
to select the best leaf extraction. Experimental results and
comparison on 152 practical soybean leaf images demonstrate
the effectiveness of the proposed algorithm for target leaf
extraction from the images with complicated background such
as with soil and residue interferents and overlaps on non-target
leaves.