Automatic recognition of scenes with power line wires in real life aerial images using DCT-based features

This paper presents results of power line scene detection methods that use new feature extraction/selection strategies based on Discrete Cosine Transform (DCT) of scenes obtained from aircraft-based cameras. Whenever a scene from an aircraft contains power lines (may that be a visible-light image or infrared), the spectrum image or DCT matrix naturally exhibits coefficients with large magnitudes. On the other hand, since the direction of cables is arbitrary, the location of the DCT extrema may appear in different positions. This work attacks the problem of extracting signatures from the DCT coefficients by systematically changing the DCT matrix sizes and applying known classifiers to the DCT sub-matrices. These sub-matrices were selected at six different sizes (4 × 4, 8 × 8, 16 × 16, 32 × 32, 64 × 64, 128 × 128) with three types of starting points: (i) top-left corner (around DC), (ii) bottom-right corner (high frequency) and (iii) block-wise scanning the complete DCT matrix. A thorough dataset that contains thousands of aerial images with cables are used for testing the efficiencies of these DCT region selection approaches. Fast and successful detection performances are obtained and presented.