This paper describes the quantitative analysis of the shape, boundary, and depth of subsurface defects by ultrasound lock-in thermography. The phase difference between defective areas and non-defective areas illustrates the qualitative analysis of the shape and the boundary of the subsurface defect. In order to accurately estimate the shape, boundary and depth of the defects, the optimal threshold value method is proposed to identify the shape and boundary of the subsurface defects based on the canny operator of image processing. A self–adaption artificial neural network (ANN) with Takagi-Sugeno modeling is proposed to determine the depth of the subsurface defect. Experimental results for a steel plate with artificial subsurface defects show good agreement with actual values.