NeuralNotwork's blog

Proposal submitted

Finished with proposal, trimmed it down, and submitted. Segmentation problems have been solved as I intend to isolate each building by tax lot. This leaves me with only the uncommon case of multiple buildings on a single tax lot, which can be solved by sorting tax lots by size and applying separate identification algorithms.

Proposal first draft finished

Unfortunately I lose at xml so this is my first draft proposal in text format. I'll clean this up tomorrow morning.

Short-term identification strategy

Refinement of the identification process will focus mainly on reduction false positives. False negatives are often easy to spot through cursory examination, and considering the current evaluation strategy (industrial standard of looking at each point with human eyes), all points are considered false negatives. False positives will have a lower chance of being discovered due to user confidence or indifference. Those structures which are identified as positive should also be marked as possible false positives if the the structure only slightly (within 20% range) meets the requirements for a p

Expected input/output

The expected input to the finished(trained) program:

Image File
The image file should be an aerial photograph of the area in question
Export ID
the export ID will identify the GIS program to which the output should be associated with
Scaling factor
the scaling factor will be the intended scale of the photo
Set of coordinates
the coordinates of the upper left-hand corner of the photo

The expected output:

New image file
This may or may not be necessary, depending upon the export ID. This file is an exact copy of the input photo, but with either a point location or an outline around target areas.
object table
This is essentially an ID-dependent file which can be read by the target application to recreate the results of the image processor.

My intended open source project

I intend to create a digital image processor which is capable of locating specific structures from aerial photographs. Exact location and accuracy are NOT short term goals of this project, instead, the identification of obviously impervious areas will be stressed. Forecasted problems with this project include:

Discerning structures covered by trees
Many structures will have some tree cover. Hopefully this can be solved by requiring polygons and arcs of a reasonable size to be connected via different traces along what is supposedly the same structure.
Syndicate content