Ms_segmenter
所属分类:图形图像处理
开发工具:matlab
文件大小:34KB
下载次数:1048
上传日期:2008-03-18 22:13:49
上 传 者:
lzj2002
说明: mean-shift用于图像分割 Matlab程序
(mean-shift procedure for image segmentation Matlab)
文件列表:
Ms_segmenter\Biomed_demo.m (52, 1998-12-12)
Ms_segmenter\Color_segment_svd1.m (2498, 1998-05-12)
Ms_segmenter\Color_segment_svd2.m (2925, 1998-05-15)
Ms_segmenter\Color_segment_svd3.m (3400, 1998-05-15)
Ms_segmenter\Cut_boundary_reg.m (1492, 1998-05-07)
Ms_segmenter\Display_Boundary.m (328, 1998-05-06)
Ms_segmenter\Display_Place_at.m (113, 1998-04-25)
Ms_segmenter\Draw_rect.m (507, 1998-05-14)
Ms_segmenter\Extract_region_reg.m (737, 1998-04-27)
Ms_segmenter\Fit_spline.m (1031, 1998-03-24)
Ms_segmenter\Full_graph.m (68, 1998-04-25)
Ms_segmenter\Gravitat.m (638, 1998-04-06)
Ms_segmenter\Gray_segment.m (2219, 2007-05-21)
Ms_segmenter\Merge_regions.m (1220, 1998-05-07)
Ms_segmenter\My_inputdlg.m (8563, 1998-04-29)
Ms_segmenter\M_shift1.m (1449, 1998-12-08)
Ms_segmenter\M_shift2.m (1916, 1998-12-08)
Ms_segmenter\M_shift3.m (1499, 1998-04-15)
Ms_segmenter\M_shift3_sps.m (1628, 1998-04-15)
Ms_segmenter\Neighbor.m (820, 1998-05-06)
Ms_segmenter\Operation_prompt.m (1454, 1998-12-15)
Ms_segmenter\Operation_prompt_right.m (1442, 1998-12-15)
Ms_segmenter\Place_at.m (417, 1998-04-23)
Ms_segmenter\Place_images.m (393, 1998-12-01)
Ms_segmenter\Region_matrix.m (473, 1998-05-07)
Ms_segmenter\Rgb_to_luv.m (1421, 1998-05-12)
Ms_segmenter\Right_button.m (476, 1998-12-15)
Ms_segmenter\Segment_demo.m (53997, 1998-12-15)
Ms_segmenter\Smooth_wv.m (893, 2007-05-21)
Ms_segmenter\Svd_dist_reg.m (3025, 1998-12-01)
Ms_segmenter\Vertex_at.m (417, 1998-04-23)
Ms_segmenter\Gray_segment.asv (2119, 2007-07-03)
Ms_segmenter (0, 2007-10-15)
Some directions and helpful hints on how to work with the region extraction/
shape classification system.
NEW FEATURES:
- region comparison now gives additional info, i.e. ratios of areas,
volumes, and average intensities.
- pixels are now removed according to regions, not to colors.
- labeling of the image is now possible. Please read below under "place a label"
- it is possible to change colors of regions ("Region" menu, "Color" submenu).
- it is now possible to pick regions not from their lists but by simply
clicking and using "(un)select 1-st/2-nd region". This is equivalent to
using the pop-up menus which remain.
- lines with which the region boundaries are drawn could be made thinner
and thicker ("Region" menu, "Thin next line"/"Thicken next line" submenus).
- images are now saved in "low" and "high" quality formats - with different
resolutions and sizes of files.
- To start the system: type in "Biomed_demo" at Matlab prompt.
- To read in an image: "Image" menu "Open" submenu ("Alt-I" and "Alt-O", or
"Ctrl-O") either in the start window or in any window containing an image.
In any case the image will appear in a separate window, not overwriting an
existing image. Currently only "tiff", "jpeg", and "bmp" formats are supported.
It is preferrable to crop the part of the image you will work with first in
some other editor, as Matlab seems to take much memory to process it when
reading, even without any processing.
- To smooth an opened image: "Image" menu "Smooth" submenu ("Alt-I" and
"Alt-M", or "Ctrl-M"). Available filters are: "Circle-5" - averaging over a
"circular" window of diameter 5 (a matrix of size 5 by 5 is taken first, 3
corner elements in each of the 4 corners are set to 0, and all others are set
to 1/(sum of elts)); "Symlet" - each color component is smoothed by means of
the particular wavelet family in the usual fashion; "Avg-3" - square 3x3
matrix of 1/(sum of elts) is taken as a filter). I prefer the first filter as
it provides quick and sufficiently good results. The wavelet-based one seems
to need some debugging (as of 11/25/***).
- To crop a part of an image: "Image" menu "Crop" submenu ("Alt-I" and
"Alt-C", or "Ctrl-C"). Then pick a vertex of the future rectangular window
by clicking once on the image with the mouse and select a rectangle that will
become a new image. When rectangle is completed a prompt whether to crop or
not appears. If "Yes" is pressed, the cropped part appears in a separate
window, while blue rectangle in the original image indicates where the part
came from. To get rid of the rectangle use "Image"/"Refresh".
- To refresh an image (get rid of various extraneous dots and lines): "Image"
menu "Refresh" submenu ("Alt-I" and "Alt-E", or "Ctrl-E").
- (NEW) To place a label on the image click on the place where the label
should begin first. Then go to "Image" menu "Label" submenu ("Alt-I" and
"Alt-L", or "Ctrl-L"). A menu offering the title for the label, the font
size, type, and color will pop up. When the label is entered, it will be
first tentatively placed on the image upon which a confirmation dialogue will
pop up. If "No" is chosen the label will not be placed. To edit an existing
label, click with the mouse close to the starting point of the label (leftmost
character) and again select "Label" submenu of "Image" menu. To erase an
existing label this step can be followed - entering in the editing mode first
and blanking the label second.
- To save an image: "Image" menu "Save as" submenu ("Alt-I" and "Alt-S", or
"Ctrl-S"). Currently only "tiff", "jpeg", "bmp", "eps" formats are supported.
The figures are saved along with all the lines they may contain. The saved
figures are usually stored in two files ("file_low.xxx" and "file_high.xxx")
where the first file is of low quality and the latter file (of usually
much bigger size) is of high quality. "Eps" is the Encapsulated PostScript
and there is just one version of it. For "bmp" the low-quality image is
placed in the file "file.bmp" while the high quality one into "file.jpg".
- To segment an image (MAIN OPERATION): "Image" menu "Segment" submenu
("Alt-I" and "Alt-G", or "Ctrl-G"). Segmentation means splitting the image
pixels into color clusters first (as prescribed by the "Mean Shift Algorithm")
and into connected regions second (as is usually done in case of images).
The segmentation parameters (appear in the parameter setting window):
- Window size: determines the radius of the search window used by the
segmentation algorithm. The smaller the window size the more likely the
search for the maximum density location will end up in a local maximum, and
therefore the bigger number of initial color clusters will be. It also
affects the value of the "cluster diameter", i.e. the value by which
two colors of the same color cluster could differ.
- Color components to use: determines the number of "principal components"
to use when segmenting the image. Since a color image consists of 3 color
components in any color system, there is a possibility to use 1-3 principal
ones for segmentation purpose. When 1 component is used, it is equivalent
to using just intensity info, while 3 components uses full color info.
The fewer color components are used, the faster the segmentation is.
The parameter is automatically set to 1 in case of grey-scale images.
- Color group size: determines the minimum number of pixels in a color
group for it to be considered a "color cluster". This parameter usually has
little effect on the final segmentation due to the fact that the number
of pixels of nearly the same color is fairly high in a typical image.
- Min region size: determines the minimum number of 8-connected pixels
to be considered as a separate connected region. Thus to find only
sufficiently big and regular regions (as opposed to small specks) the
number must be set to a sufficiently high value.
- Separate layers: determines whether to show ("yes") or not ("no") all
color clusters at once as "layers" of the original image, i.e. nearly all
the image pixels are placed into one of the layers. The white color of a
particular pixel in the produced black and white figures is interpreted
as "the particular pixels is placed in the particular layer", and vice
versa. Some pixels not belonging to any color clusters (due to
their color deviation) will not belong to any of the layers.
NOTE: the regions handling which will now follow is based on the layers.
Having segmented the image (the operation may be repeated if necessary) we
now can apply various operations to the resulting segments. Segments are
always connected (unless obtained by explicit merging of two disconnected
ones).
- To find a region's boundary: click on a point belonging to the desired region,
then "Region" menu, "Find boundary" submenu ("Alt-R" and "Alt-F", or "Ctrl-F").
NOTE: the following operations will require selecting one or two regions. The
selection window is basically the same for all the operations. A list of
regions (with their numbers) is presented. Upon (tentatively) selecting a
region the region's color changes from a non-white to white. Upon deselecting
a region the region's color is changed back from white to a non-white.
Selecting is done by clicking on the region's number in the list (which should
hang down when the down arrow in right side is pressed). Deselecting is done
automatically when another region is selected or "cancel" button is pressed.
NEW: now region selection can be done by clicking on the region and choosing
"(un)select 1-st/2-nd region" from "Region" menu (or "Ctrl-1"/"Ctrl-2").
The region chosen again becomes white and turns its usual color when another
region is selected. For most operations (e.g. Region save/Cut/Intensity 3-d)
choose the region to be the 1-st. The second region needs to be chosen only
for binary operations, e.g. comparison or merge. Thus to compare two regions
in the same window one would click/select 1-st and 2-nd region and then choose
"Compare...Image regions". To compare two regions in different window one
however needs to pick 1-st region in one window and select
"Compare...Window regions" in that window. Then 2-nd region needs to be
chosen in the second window and the same menu item "Compare...Window regions"
needs to be selected, so that the system would know which two 1-st/2-nd
pair of regions to compare.
- To save a region (in a format that can be later used for shape comparison
with other regions): "Region" menu, "Save as" submenu ("Alt-R" and "Alt-V",
or "Ctrl-V"). The files have extension ".rgn" (for "ReGioN").
- To compare two regions in terms of their shapes (as described in the
article) and obtain some additional info:
"Compare to" submenu of "Region menu". A region can be compared with a
region of the same image ("Image region" submenu of "Compare to" submenu
("Alt-R" and "Alt-C" and "Alt-I", or "Ctrl-I"), or a region in a
different window ("Window region" submenu of "Compare to" submenu ("Alt-R"
and "Alt-C" and "Alt-W", or "Ctrl-W"), or a region previously saved on disk
("Disk region" submenu of "Compare to" submenu ("Alt-R" and "Alt-C" and
"Alt-D", or "Ctrl-D"). In the first case the choice of the two regions is
done in the same dialogue. In the second case the action of choosing the
regions must be done in the two windows (same sequence of keys must be
repeated in both windows, and windows may coincide). In the third case after
the image region is chosen a prompt to choose a saved region is given.
- To merge two regions (which may be disconnected): "Merge" submenu of
"Region" menu ("Alt-R" and "Alt-R", or "Ctrl-R"). The region is then treated
as a single entity, while the regions from which it was formed "disappear"
from various menus (e.g. "Compare", "Merge", etc).
- To cut a region into two pieces: "Cut" submenu of "Region" menu ("Alt-R"
and "Alt-U", or "Ctrl-U"). Upon choosing the region a prompt to draw the
separating line appears. Please be very careful drawing the line as the
algorithm for separating the region is quite unstable and must be improved.
The line must start and end a little bit outside the region. Upon successful
splitting 2 new regions will appear while the old one will disappear.
- To visualize a region and obtain some basic info: "Intensity 3-d" submenu
of "Region" menu ("Alt-R" and "Alt-3", or "Ctrl-3"). As of today (11/29/***)
the region is shown correctly, not flipped. In the image the horizontal axes
correspond to image dimensions, while the vertical one corresponds to image
intensity. The 3-dimensional graph can be rotated left and right, and up and
down by means of "Ctrl-L", "Ctrl-R", "Ctrl-U", and "Ctrl-D" keys while the
window is active. The amount by which the graph is rotated could be
increased/decreased by means of "Ctrl-A" and "Ctrl-V" keys. While these
operations don't allow to choose an arbitrary point of view, large flexibility
is achieved. To obtain some relevant to the region information use "Values"
menu of the graph window.
- To exclude some regions from further consideration (will be shown usefulness
of later): "Background" submenu of "Region" menu ("Alt-R" and "Alt-K",
or "Ctrl-K"). Now the pixels are removed according to the found regions,
not underlying colors. Thus the background should preferrably consist of
one or few regions.
- To change the color of a region's boundary, first select the region, and
second click on "Region"/"Color" submenu (or "Ctrl-L"). A list of possible
colors to choose from will be presented.
- To thicken/thin the next boundary use "Thicken/Thin next line" submenu of
"Region" menu. This could be useful if the lines are too thin for printing
purposes. If the next line is still not thick enough, try to do it several
times.
A typical use of the system:
- read in an image ("jpeg", "tiff", "bmp" formats only) (Image/Open)
- select a rectangular region to work with (Image/Crop)
- segment with the initially given parameters (Image/Segment)
- if the resulting regions are too small (oversegmentation), increase the
window size in Segmentation parameters window and resegment;
- if the resulting regions are too large (undersegmentation), reduce the
window size in Segmentation parameters window and resegment;
- if there is large uniformity in "the background":
- first segment the image with a small window size but large enough
to capture most of the background into one or few regions;
- second use "Background" operation from "Region" menu to exclude the
corresponding regions from further processing (pixel removal);
- third resegment the image with a larger window size so that to capture
less uniform foreground;
- if a region is best obtained as a result of merge of two fairly different
(in color respect) regions that are adjacent, use "Region/Merge" operation.
- if a region has some extraneous parts (may be from other similar regions
as in case of cells), use "Region/Cut" operation.
- having obtained a suitable region it can be visualized in terms of its
intensity (Region/Intensity-3d).
- having obtained two regions (in the same window or different ones) they can
be compared in terms of their shapes and some other respects (e.g. areas or
average intensities) (Region/Compare to/[Disk/Window/Region])
- an exemplary region can be saved (shape info only, no intensity info) for
future comparisons.
- an image with or without lines could be saved for future work (Image/Save).
近期下载者:
相关文件:
收藏者: