You are here: Technology Digitization Space imagery decoding

Space imagery decoding

Rate this item
(6 votes)

If the task is more complicated than area calculation at oil spillage or estimation of burnt wood, one may be sure that photo decoding will be mostly done manually.

Lately we started gradually "to stalk up" to some problems of space imagery decoding. In the course of our work for the Rapid Eye Company we had to recognize some areas basing on a series of criteria and considering FOUR photos at once.

Strangely enough, most objects were extracted applying a trivial operation series Diffuse - Contrast Enhancement- Brightness-based Filtering. After it, the objects were automatically outlined. But in a number of cases groups or chains of small objects were extracted instead of big spots. Iimprovements made in the program for their quick assembling are considered below.

Well, it remains up to you to select accuracy of object positioning - Easy Trace provides all necessary means for it. And the speed of decoding is shown in this video: Video: Imagery decoding for address plans making.

The video is based on an actual work record. All operations of scaling, navigation in the photo, and object forming are shown in real time. Some new features of ET v. 9.0 are used at the work.

Broken Line Tracer

One often faces with the dilemma at data extraction from imagery (vectorizing of roads, etc.). Road lines have long straight segments as a rule as well as some points of inflection. On the one hand, these bends are better seen "from afar". On the other hand, vector lines should coincide with the center of raster lines, i.e. you need to "move the image near". As a result, vectorizing of roads either turns into a tiresome sequence of zooming in and out or gives numerous unnecessary vertices in straight line parts.

Automatic zoom at vertex placing is a solving of this problem. It resembles dynamic scaling at point insertion provided for the Point tool. Specify an approximate position of the next vertex? Press the mouse left button but do nor release it. Easy Trace automatically zooms in the working window. Move the cursor to the exact position of the vertex and release the button there. The vertex will be added and the scale restored automatically.

You may fix this temporary zoom with the Ctrl + Shift combination, input a series of vertices (a road bend) and then press Ctrl + Shift again to restore the scale, i.e. to "move the image away".

Rectangle tool

Some innovations introduced in the Rectangle tool made it more effective at building decoding during address plan making.
  • Address plans often contain numerous buildings of the same size and orientation relative the street line. Vectorize one such rectangle, then click the mouse right button inside it to capture the sample and start cloning. One left click is sufficient now to place the next object (rectangle).
  • Neighboring buildings are often perpendicular to each other. When in the cloning mode, a right click inside an existing rectangle or on a linear object causes orientation of the next clone at right angles to the specified object. Subsequent right clicks rotate the clone cyclically by 90 degrees. Thus, any preexisting object may be used for quick orientation of the next rectangle (building).
  • You may also start clone rotation and input a rectangle at an angle different from 90 degrees. It will take two clicks to specify position of the rectangle corner and to fix its slope angle.
  • To change dimensions of the next clone, use the mouse wheel at Ctrl (for the long edge) or Shift (for the short edge) key pressed.
  • You may specify temporary guides at input of the first rectangle in a group and to align all the following objects along them - new-added rectangles will be pulled to the guides automatically. This feature makes input of aligned one-type building contours even faster.
Tool control is very simple - the Ctrl + Shift combination activates cyclically the modes: "Cloning" - "Clone rotation" - "Guide forming". Capture of new patterns and size change with the mouse wheel (at Shift or Ctrl pressed) are accessible in any of the modes.

Spline mode of the Curvilinear Tracer

This mode is intended for quick and accurate forming of complex curves. It is useful when you vectorize boundaries of natural or complex artificial objects in space imageries.

Interface of the tool differs markedly from one of similar tools in other applications. It is characterized by simple and univocal line form control with the help of mouse cursor move. The tracer creates a curve corresponding to the form of an object in the image rather than an abstract smooth curve through specified points.

Spline mode is a submode of semiautomatic tracing that may be started when necessary. For example, you may begin semiautomatic vectorizing of a curve, then create a long straight segment manually and complete the operation by forming of a complex curve in the spline mode. Final result is a common polyline.

Tracing control in the spline mode differs a little from usual but remains simple. The first click in this mode defines a point the curve should pass through (usually, it is the vertex of a curve bend) and the second fixes the line part. Move of the cursor forms the bend. In difficult cases press Ctrl and use the mouse wheel to form it more precisely.

It is also possible to create Custom tools that would start the Curvilinear tracer in the spline mode from the beginning.

Enclose tool

The Enclose tool received a new option - contour smoothing with a user-specified approximation. It is provided for quick forming of encircling contours around groups of selected vector objects. Final shape improvement of the contours was done with the Camber Editor.

Synchronize and Split view mode

The Synchronize and Split view mode divides the working window into 2 or 4 parts, and you may select the visible image individually in each of them. Project positioning and scale in the split window are synchronized - all the parts always contain the same zone of the project.

Vector editing and navigation in the image may be done in any part of the window. The mode is extremely time-saving if several images are required simultaneously to control vector data.

Transparency mode

Usually, two images are sufficient for vector data control. For example, that is true for digitization of a color map with a superimposed color photo. It is enough then to use transparent image superposition instead of the Synchronize and Split view mode.

Transparent presentation of color images through black-and-white ones was first provided in Easy Trace long time ago. It is convenient at semi-automatic digitization. But comparison of two full-color images (a map and a photo or two photos made at different time) requires rigorous control of transparency. A new video-mode for concurrent presentation of full-color images is called Image Transparency.

.

Road network recognition

Recognition of the road network is a typical task at imagery decoding. Regular rows of city blocks in photos just put the idea of automatization into one's mind. For example, it is possible to input points at road crossings applying the dynamic scaling mode of the Point tool and then somehow link them with lines.

This idea was realized in the new Generate Grid from Points utility. You just put points at nodes, start the utility, and the network is ready. It will have some false segments most likely and some others will require editing but on the whole it accelerates and simplifies the process notably.

One may run the Buffer Zone Generation utility after network forming to get polygonal coverage of roads.

By the way, agricultural grounds in foreign photos often look like neat sets of rectangles separated by country roads. As a matter of fact, this utility was developed for them.

Multispectral images

...If Easy Trace is to process imageries indeed, it should learn how to make composites "in passing". Water is distinctly visible in one composite, woods in another, swamps in yet another and so on.  So, it would be reasonable to make ready pre-sets. "Select satellite type - Landsat? QuickBird?" – selection of processing strategy. "What do you want to recognize? Woods? Fields? Water? Soils?" – selection of the corresponding composite.
Plus manual composite forming for advanced users.
voffka

This topic if our forum has prompted us to develop a new Easy Trace Pro feature. Loading and viewing of multispectral images saved as one TIFF – file are realized. Two modes of image viewing is provided: as a RGB-composite of selected ranges or as a grayscale representation of the selected range.

Any current representation of a multispectral image may be saved as a common image for further processing.

P.S.

We are novice in imagery decoding. Suggestions and advices received from our users were previously extremely useful for software development but gradually we have run out of them. Most of urgent problems were solved and in some issues we started to outdistance traditional approaches. But now we are at the starting line again and we shall appreciate your ideas that will hopefully make our program as skilled in imagery decoding as it is in digitization.

More in this category: « Prev Next »

Add comment


Security code
Refresh