Two types of basic operators able to extract linear elements can be usually observed:

• detection operators, and
• tracking operators.

In recent research, can be noticed a tendency to integrate various types of operators in order to make them cooperate. This leads to more robust methods which provide more reliable results.

Thanks to the increase in the performances of computers, research is directed towards new operators, more complex but also closer to human perception, such as neural networks, fuzzy reasoning, etc.

Detection operators:

They are applied globally on all the image, pixel by pixel. Decisions are taken independently from each other by analyzing the immediate neighbourhood of each pixel.

These method are often sensitive with noise and thus very related to the nature of image. An accurate adaptation of the operator is difficult and leads generally to an over-segmentation, or to an under-segmentation. Moreover they do not always provide perfectly connected chains. The pixels selected constitute only ``seed'' points (``seeds''), and it is necessary to use aggregation operators to link the pixels in chains.

Tracking operators:

They guarantee obtaining connected chains, because they work step by step by selecting the best candidates to increase the chain. Sensitivity to the noise is less important since these operators use a knowledge ``a priori'' which they progressively refine as the construction of the chain goes along.

But two disadvantages must be noticed about these processes:

• it is necessary to detect ``seed'' points to initialize the tracking, generating uncertainties, and
• if an occlusion occurs in the lineament, the process stops.
  
  

Aggregation operators:

They are used after the first detection carried out by the operators mentioned above. They try to group and connect, in a coherent way, the adjacent elementary primitives according to their similarity.

In the case of lineaments, it is a matter of connecting end to end the segments to form roads or rivers.

• Detection Operators
• Thresholding
• Edge Detection
• ISEF - Infinite Size Symmetric Exponential Filter
• Line Detection
• ``Top Hat'' Operator
• VANDERBURG Operator
• DUDA's Roads Detectors: DRO, MDRO and EMDRO
• DRO: "DUDA Road Operator"
• MDRO: "Modified DUDA Road Operator"
• EMDRO: "Extended Modified DUDA Road Operator"
  
  
• Tracking Operators
• Structural Tracking
• Dynamic Programming
• A* Algorithm
• F* Algorithm
• Automatic Tracking using Dynamic Programming
• Theory
• Applications
• Active Edges
  
  
• Aggregation Operators
• Contextual Grouping
• Perceptual Grouping
• Lineaments Connection
  
  
• Application: Geographic Features Extraction by use of Satellite Images and Vectored Imprecise Databases
• Principle and Description
• Detection
• Fusion
• Optimization
• Smoothing
• Application and Results
• Early Test
• Gafsa - Tunisia: a Small Road
• Toulouse - France: Canal du Midi
• Toulouse - France: South Expressway
• Conclusion