Level 2 Help for HORIZON

A single 1-band grayscale input image. The program works better if the input 
image is radiometric corrected.

The output of this program will be a Comma-separated values (CSV) file.
The length of the output CSV file should always be exactly the the same as
input image's width. Each integer value in the CSV file represents the
horizon border's LINE position in that specific SAMP.
For example:
  324,325,325,326,327, ... ...
This means that at column or SAMP 0, the horizon border is at row or LINE 324;
at column or SAMP 1, the horizon border is at row or LINE 325, and so on. 

Minimum threshold value for detecting strong horizon borders. The program will
iteratively search through gradient magnitude image for finding preliminary 
horizon borders. This parameter defines at what value the search starts. 
According to the definition of Sobel operator, the pixel intensity range in 
gradient magnitude image is from 0 to 2.68E+8. See METHOD section for how 
the range is calculated. In theory, we should choose the minimum value to be 
0. However, if we do, the program will become very sensitive to weaker horizon 
borders and the noisy edges mostly present in sky region in the gradient magnitude 
image. The program should still be able to find the horizon borders, but it 
requires more computation power to achieve that. In practice, it is unlikely 
the strong horizon border pixel values in gradient magnitude image will be smaller 
than 200, so it is set to be 200 by default. Decreasing the threshold value should 
make the program sensitive to weaker horizon borders. 

Maximum threshold value for detecting strong horizon borders. Similar to parameter
SH_THRESH_MIN, the difference is this parameter defines at what value the search
ends. According to the definition of Sobel operator, the pixel intensity range in
gradient magnitude image is from 0 to 2.68E+8. In theory, the maximum value should
be 2.68E+8. However, for a natural 12-bit image it is unlikely the intensity 
difference will reach 4096, which means the searching process is also unlikey
will reach 2.68E+8. In practice, almost all the times, the program finds the optimal 
solution when SH_THRESH_MAX < 1000, so the default is set to be 1000. 

Strong horizon search incremental factor. This parameter should be used together with
SH_THRESH_MIN and SH_THRESH_MAX to be meaningful. The program will iteratively search
through gradient magnitude image for finding preliminary horizon borders. It defines
the sampling step from SH_THRESH_MIN to SH_THRESH_MAX. Considering the balance between
search precision and computational complexity, the default is set to be 50. Increase
the value of this parameter should cause seach precision and computational complexity
to decrease. Decrease the value of this parameter should cause search precision and
computational complexity to increase.

Minimum threshold value for detecting weak horizon borders. Similar to SH_THRESH_MIN.
The difference is this parameter is used to find weak horzion borders. Weak horizon
borders will be represented as weaker edges in gradient magnitude image. This parameter
should be set to a little bigger than 0. Doing so will make the program sensitive
enough to find weak edges. Addtional processes are enabled to exculde noisy edges when 
detecting weak horizon borders. See METHOD section for details.
 

Maximum threshold value for detecting weak horizon borders. Similar to SH_THRESH_MAX.
The difference is this parameter is used to find weak horzion borders. This parameter
is usually set to equal to or a little smaller than SH_THRESH_MIN.  
 

Weak horizon search incremental factor. Similar to SH_SIF. This parameter should be 
set to smaller than SH_SIF.

If the position a horizon border is vertically away its N immediate neighbors by the 
value of this parameter in pixels, the horizon border will be classified as an outlier. 
This parameter is used in outliers and weak horizon borders detection and refinement 
step.
 

Outlier search interval factor. This parameter defines how many immediate neighbors 
in one direction to be used for calculating the average to compare with the target 
horizon border. The total number of neighboring pixels to be used is 2 times this
parameter.

Gradient magnitude image weak horizon borders threshold value. This parameter is used 
for detecting the presence of weak horizon borders. If the current pixel value in
gradient magnitude image is greater than the average of the current pixel column by 
the value of this parameter. The current pixel will temporarily be marked as possible
weak horizon border for further verification. 

Gradient direction image weak horizon borders threshold value. This parameter is used 
for verifying if a temporary weak horizon border previously marked using parameter 
WH_MAG_THRESH is a true weak horizon border. For a temporary weak horizon border previously
marked using parameter WH_MAG_THRESH, if the absolute difference between the temporary weak 
horizon border and 1/3 of its neighboring pixels' direction defined in a N x N space is 
within the range of -WH_DIR_THRESH to WH_DIR_THRESH, then the temporary weak horizon border
will be classified as true weak horizon border.

Gradient direction image weak horizon searching interval factor. This parameter is 
used to define a N x N search space for identifying a possible weak horizon border.
In general, this parameter shouldn't be changed very often. It is provided to handle
some extreme situations. 

Maximum iterations to search for outliers and weak horizon borders. Outliers and weak 
horizon borders detection and refinement is a iterative process. The pamameter defines 
the maximum iterations the program will detect and refine ourliers and weak horizon 
borders. The maximum iteration cannot exceed 30. If it exceeds 30, it will be reset to
be 30 in the program. If this parameter is set to 0, then the outliers and weak horizon
borders detection and refinement process will be turned off.

Border average acceptance threshold value. This parameter is used together with 
BAR_THRESH, BP_FACTOR, BAADA_THRESH, and BAADR_THRESH to remove outliers and
columns without sky region. Please see the algorithm described in METHOD section
for how to use this parameter.

Border average rejection threshold value. This parameter is used together with 
BAA_THRESH, BP_THRESH, BAADA_THRESH, and BAADR_THRESH to remove outliers and 
columns without sky region. Please see the algorithm described in METHOD section
for how to use this parameter.

Border partition factor. This parameter is used together with BAA_THRESH, BAR_THRESH,
BAADA_THRESH, and BAADR_THRESH to remove outliers and columns without sky region.
The program provides the option to remove outliers and columns without sky region 
using the horizon borders as a whole. The program also provides the option to
first cut the horizon borders into parts, and then remove outliers and columns without
sky region. This parameter indicates how many parts to cut the horizon borders.

Border average of absolute difference acceptance threshold value. This parameter is 
used together with BAA_THRESH, BAR_THRESH, BP_THRESH, and BAADR_THRESH to remove 
outliers and columns without sky region. Please see the algorithm described in 
METHOD section for how to use this parameter.

Border average of absolute difference rejection threshold value. This parameter is used
together with BAA_THRESH, BAR_THRESH, BP_THRESH, and BAADA_THRESH to remove outliers
and columns without sky region. Please see the algorithm described in METHOD section
for how to use this parameter.

Optional output for smoothed image. The size of the smoothed image will be slightly 
smaller than the original image depending on the size of filter window is used. 
The width and height of the smoothed image output is given by the two equation below:

  smoothed_image_width = original_image_width - 2 * floor(filter_size / 2)
  smoothed_image_height = original_image_height - 2 * floor(filter_size / 2) 

After applying the smoothing filter on the original image, the smoothed image has the 
same width and height compared to the original image, but the pixel values of the borders 
of the smoothed image will not be accurate. Instead of keeping the inaccurate pixel values,
the program will cut the borders.

Optional output for gradient magnitude image. Gradient magnitude image is calculated using 
Sobel operator. For detailed calculation, please see METHOD section. Gradient magnitude 
image provides analytical and other useful information. It is recommanded to output the 
gradient magnitude image, especially when tweaking the parameters. 

Optional output for gradient direction image. Gradient direction image is calculated using 
Sobel operator. For detailed calculation, please see METHOD section. Gradient direction 
image is used when identifying whether a pixel or a group of pixels is weak horizon 
border(s) or noise. It is recommaned to output the gradient direction image, especially 
when tweaking the parameters.

Optional output for horizon borders overlay on top of the original image served for quick 
look purpose. The final horizon borders will be assigned with DN value 10000, and the
preliminary horizon borders ill be assigned with DN value -10000. It is highly recommanded
to output the quick look image.

The smoothing filter size in pixels. The minimum filter size allowed in this program is
3 x 3. Filter size significantly affects the program's execution time. The computational
complexity of applying a smoothing filter to an image is O(MNC^2), where C is the filter 
size, M is the image width, and N is the image height. Filter size should be defined based 
on the size of noises. In general, if the noise size is 3 x 3, then a filter size of 
7 x 7 should be effective to blend them.

The smoothing filter type. Three filter types are provided in this program. They are
MIN, MAX, and MEDIAN. MIN filter is usually used to blend salt noise. MAX filter is 
usually used to blend pepper noise. MEDIAN filter is usually used to blend salt-and-pepper
noise.

Number of line(s) to cut from the upper border of the input image. If this parameter is 
set to 0, then nothing will be cutted. 

Number of line(s) to cut from the lower border of the input image. If this parameter is
set to 0, then nothing will be cutted.

Number of sample(s) to cut from the left border of the input image. If this parameter is
set to 0, then nothing will be cutted.

Number of sample(s) to cut from the right border of the input image. If this parameter is
set to 0, then nothing will be cutted.

Optional output for image with borders trimmed.