In order to improve precision and adaptive capability of road defect detection system, an image enhancement technique for automatic road defect is investigated. To estimate the crossover points of gray transform, a new 3-D pattern formed with two local features of the neighborhood of each pixel and one global feature based fuzzy measurement is constructed for pixel classification. In design of a fuzzy membership function with image enhancement ability, the Non-uniform Rational B-Splines(NURBS) is used to produce a double S shape fuzzy membership function as the gray transform function. The gray transform function can easily be united with the crossover points perfectly and its designable shape can offer capability of gray level centralization. Therefore, the method can obtain the road crack only by a few operations of iteration and enhancement. Experimental results testify that this adaptive technique can provide satisfactory enhancement effects and the detection accuracy of road crack region pixels reaches 95%. The proposed technique shows better reliability and precision for road defect detection systems.