Programmig Language: C++/CLR

Determination of each object using the K-Means algorithm on an image and the result of the feature extraction of each determined object is the object detection. First, an entrance image is taken and Trained. After that, "db.txt" is used for the Test process and the classification process is done.

• K-Means algorithm was used for Unsupervised Learning Classification.
• Euclidean or Mahalanobis distance calculation methods can be used for the K-means algorithm.
• For noise reduction :
  • Mean Filter
  • Morphology -> Opening ( Erosion/Dilation )
• Component Connectivity Analysis process is done with 8 Neighbors (Structure Element).
• Feature extraction, 7-level moment is calculated.

1. Apply Mean Filter to input image
2. KMeans algorithm apply
3. The background of the result image is set to 0; make foreground 1 (It's now binary image)
4. Morphology steps:
   • 4.1. Apply Erosion on Binary image
   • 4.2. Apply Dilation and Opening process completed.
5. Object labeling is done with the Component Connectivity Analysis process.
6. Bounding boxes are apply to labelled images (drawRectangle)
7. TRAINING Steps:
   • 7.1. Get object name from ** TextBox **
   • 7.2. Perform Feature Extraction for each object enclosed in Bounding boxes. (7-level moment)
   • 7.3. Save in "db.txt"
8. TEST Steps:
   • 8.2. Get moment values and object names in "db.txt"
   • 8.3. For each detected object:
     • 8.3.1. Apply Feature extraction process
     • 8.3.2. Match the values in "db.txt"
     • 8.3.3. Add the most similar objects to "RichTextBox" with their own color

Sample 1 is lentil test

Sample 2 is lentil train

Sample 3 is many objects detection with Euclidean Distance

Sample 4 is many objects detection with Mahalonobis Distance