The detection of man made objects in images is a complex problem in the image processing area which has many applications.
The detection of man made objects in images is a complex problem in the image processing area which has many applications. There are many different approaches to face the problem. In this project, we will present an algorithm based on multi-channel filtering using Gabor filters for distinguishing between man-made objects and natural background by creating an output image in which the man-made objects are marked. The algorithm will be tested on monochromatic and multispectral satellite images.
Scanning and detecting man-made objects localized in a natural scenery is a difficult task which could be solved mainly today by searching them using the human vision system. The project goal is to create an algorithm for automated scanning and detection that can help us decide which image is suspected of containing man made objects and therefore requires further examination. The more we increase the detection ability and decrease the false alarm level, the more we save on human effort.
The Detection Algorithm
The detection algorithm for monochromatic images is implemented according to the next four steps:
1) Filtering the input image using the gabor filters
2) Post processing of all the filtered images
3) Applying a thershold for marking suspected pixles
4) Morphological opreations for marking the man made objects
For the multispectral images we will use the mentioned steps for each band and then combine the marked images into one output image.
Block diagram of the algorithm:
Figure 1 – Block Diagram of the algorithm
Figure 2 – Marked Image 1
The results are a good starting point for more research on scanning and detecting algorithm using the gabor filters. The detection quality is very high, but the false alarm level is quite high in certain images. in general, the false alarm and detection level depend on the parameters, chosen experimantlly by us. The implementation of the algorithm using Matlab is very efficient for it is based mainly on linear filtering.
- Yiming Ji, Kai H. Chang, Chi-Chang Hung, “Efficient Edge Detection and Object Segmentation Using Gabor Filters”
- Anil K. Jain, Farshid Farrokhnia, “Unsupervised Texture Segmentation Using Gabor Filters”
- Khaled Hammouda, Prof. Ed Jernigan, “Texture Segmentation Using Gabor Filters”
- Peter de Rivaz and Nick Kingsbury, “Complex Wavelet Features for Fast Texture Image Retrieval”
- Dengsheng Zhang, Aylwin Wong, Maria Indrawan, Guojun Lu, “Content-based Image Retrieval Using Gabor Texture Features”