SURVEY GOALS

SEGMENTATION: WATERSHED SEGMENTATION

The intuitive idea of the watershed segmentation comes from topography field.
The image is considered as a landscape relief, and heights of the different points are associated
with grey levels. Assuming that the local minima [1] are pierced with holes and the landscape is
immersed in a lake, the water filling up these minima generates the catchment basins, whereas
watershed lines are the dams built to prevent mixing waters coming from different basins. Thereby
the stochastic watershed result depends on two parameters, the number of markers N that
determines the number of detected regions and the number of realizations M. Therefore this
empirical choice performed by theuser controls the segmentation result.
It introduces a multipass fast watershed-based method (MPFW) [2] to segment both
nucleus and cytoplasm from large cell masses of overlapping cervical cells in three watershed
passes. The first pass locates the nuclei with barrier-based watershed on the gradient-based edge
map of a pre-processed image. The next pass segments the isolated, touching, and partially
overlapping cells with a watershed transform adapted to the cell shape and location.
This method based on watershed and active contour models were proposed for the
whole heart CT image segmentation task. Firstly, the Bilateral filtering approach is used to
eliminate the noise of the cardiac CT images. Then, the watershed segmentation [3] method is
applied to determine the initial seed contours. Finally, the active contour model is adopted in the
evolve process iteratively with the previous contour to obtain the precise segmentation boundaries
for whole heart CT images. The performance is evaluated with a clinical cardiac CT images
dataset, which proved good segmentation results as well as the high robustness.
New scheme of process to applying clustering technique to produce the primary
segmentation [4] before applying the improved watershed algorithm. The uses of k means
algorithms which is unsupervised learning algorithms will help watershed segmentation to
improve segmentation process. Then, the improved watershed segmentation used the level set
thresholding based on gradient magnitude map to reduce the number of false edges and over
segmentations. This proposed method demonstrates a significant improvement that may serves as
a computer aided tools for radiologists in detecting microbleeds in MRI images.
The flooding with pruning helps in estimating the boundary of an object and
assesses the regions of segmented image. Further, the objects of an image are processed with the
morphological operations like erosion and dilation to improve the accuracy of the proposed
method. The over-segmentation [5] can be avoided by recognizing the set of pixels, which exhibit
the properties of segmentation. The properties of segmentation include flooding and receiving the
flooded response with the neighboring pixels of an image. The Proposed method has been designed
with an improved watershed algorithm to assess the segmented objects and to boost.
 It improve the watershed algorithm based on the bi-dimensional ensemble empirical
mode decomposition (BEEMD), and apply the improved algorithm to the brain MR image
segmentation. In this paper, the BEEMD [6] is used to decompose the MR images into multiple
BIMFs with different frequency. For different segmented objects, these BIMFs are given different
weights to enhance the details of the various regions. Then, the enhanced image is segmented by
applying improved watershed algorithm. Finally, correct the segmentation results and fix their
blurring details. This method improves the over-segmentation problem of watershed and the details
blurred problem of the improved watershed.
The algorithm takes Five module method (FMM) [7] compressed retinal images as
the input. Watershed lines and canny detectors are used to find the vessel like candidate segment.
Different features namely shape of the segment, position of the segment from the origin,
positioning, intensity of the segment in the image, divergence, and line density are extracted for
each candidate segment. Each candidate segment is labeled as normal or abnormal based on its
features using Support Vector Machine (SVM) classifier.
Novel brain tumor segmentation scheme using fractional order sobel mask and marker
controlled watershed transform is proposed. To obtain the bright tumor region [8], regional
maxima operation is performed on morphological preprocessed input T2-weighted MR image. The
output regional maxima image is taken as an internal marker. Distance transform based watershed
transform is applied on regional maxima image, the watershed ridge lines are used as external
marker. Now, the fractional sobel mask of order is 0.3 is applied on input T2-weighted MR brain
image to obtain gradient magnitude image.
Learned boundary maps [9] are known to outperform handcrafted ones as a basis
for the watershed algorithm. We show, for the first time, how to train watershed computation
jointly with boundary map prediction. The estimator for the merging priorities is cast as a neural
network that is convolution (over space) and recurrent (over iterations). The latter allows learning
of complex shape priors.
This disease using mammography technique, segmentation is an important step to
Detect the suspicious region(s) of mammograms. Segmentation concerns to the process of division
of mammograms into different sections. Objective of segmentation [10] is to simply modify the
presentation of an image so that it becomes more significant and easier to study. Although many
algorithms have been proposed yet to segment out the suspicious regions of mammograms,
automatic segmentation of masses of improved quality is still considered to be difficult. This study
introduces a novel marker controlled watershed algorithm for segmentation of mammograms to
highlight the suspicious regions.
Computer vision [11] occupies an important position in the recognition of the
image processing, in order to accurately get the grape disease and the degree of damage. An image
segmentation technique is proposed to improve the morphological watershed algorithm, and edge
detection and threshold segmentation are used to realize the image in the normal parts and diseased
parts of the grape segmentation.
 Most contemporary approaches [12] to instance segmentation use complex
pipelines involving conditional random fields, recurrent neural networks, object proposals, or
template matching schemes. In this paper, simple yet powerful end-to-end convolutional neural
network to tackle this task. This approach combines intuitions from the classical watershed
transform and modern deep learning to produce an energy map of the image where object instances
are unambiguously represented as energy basins. And then perform a cut at a single energy level
to directly yield connected components corresponding to object instances.
Delineating individual tree crowns (ITCs) in highspatial- resolution images can
help to improve forest inventory and management. However, single-band watershed segmentation
methods [13] often fail to delineate broadleaf species, particularly in unevenaged stands when a
single-scale parameter is used to fit segments to reference crowns of different sizes. In this study,
we present multispectral watershed segmentation and multiscale fitting method for ITC
delineation, and the method involves two steps: 1) multispectral watershed segmentation to
produce multiscale segmentation for subsequent fitting, which takes full advantage of boundary
information contained in the spectral contrast of multiple bands; and 2) multiscale fitting to
identify optimal parameters to best fit each ITC, rather than selecting a single parameter value
based units overall fit to the image as a whole.
Multipass fast watershed-based method (MPFW) to segment both nucleus and
cytoplasm from large cell masses of overlapping cervical cells in three watershed passes. The first
pass locates the nuclei with barrier-based watershed on the gradient-based edge map [14] of a pre-
processed image. The next pass segments the isolated, touching, and partially overlapping cells
with a watershed transform adapted to the cell shape and location. The final pass introduces mutual
iterative watersheds separately applied to each nucleus in the largely overlapping clusters to
estimate the cell shape. In MPFW, the line-shaped contours of the watershed cells are deformed
with ellipse fitting and contour 0adjustment to give a better representation of cell shapes. The
performance of the proposed method has been evaluated using synthetic, real extended depth-of-
field, and multi-layers cervical cytology images provided by the first and second overlapping
cervical cytology image segmentation challenges in ISBI 2014
and ISBI 2015.
Practical features of hierarchies of morphological segmentations, namely the
quasi-flat zones [15] hierarchy and watershed hierarchies, and to evaluate their potential in the
context of natural image analysis. A novel evaluation framework for the hierarchies of partitions
designed to capture various aspects of those representations: precision of their regions and
contours, possibility to extract high quality horizontal cuts and optimal non-horizontal cuts for
image segmentation, and the ease of finding a set of regions representing a semantic object. This
framework is used to assess and to optimize hierarchies with respect to the possible pre- and post-
processing steps.