Abstract: Investigates detection of small targets in complex natural background by way of improving Kaplan's extended fractal feature to adapt itself to target dimensions to a certain extent. In view of the natural background and both the horizontal and vertical directionalities of the speckled man-made targets, a multi-scale extended fractal feature via wavelet analysis is proposed utilizing the directionalities which are well decomposed by wavelet to combine the horizontal and vertical wavelet decomposition of their mutual energy respectively, thus intensifying effectively the target image both horizontally and vertically with background disturbance suppressed. Experimental results showed that the target detection based on such a feature is stable and adaptable to infrared/TV images on either complex ground or sea surface and available to locate quickly the small target from single-frame images.