Combining three-dimensional (3D) imaging ability of optical coherence tomography (OCT) with movement recognition ability of dynamic scattering technique, label-free 3D OCT angiography can be realized, which has a wide range of applications in basic science research and clinical diagnosis. At no expense of line scanning speed, the scale of capillaries can be detected by improving the sensitivity through the interframe analysis. However, there exists a certain residual overlap between dynamic flow signals and static tissue beds due to a series of reasons, thus making it difficult to completely distinguish dynamic flow signals from static tissue beds. Thus, when it comes to threshold segmentation for the blood flow signal extraction, classification error rate is inevitable, resulting in the decrease of the motion contrast of angiogram. In order to reduce classification error rate between static tissue beds and dynamic flow signals for high motion-contrast angiography, we propose a method of component compounding in wavelet domain. Three main steps are needed for this method. Firstly, on the basis of two-dimensional (2D) discrete static wavelet transform, a frame image can be decomposed into multiple levels. Each level has four components, i.e., approximation component, horizontal detail component, vertical detail component and diagonal detail component. Different decomposition levels and types of wavelet can be selected according to the demand. Secondly, the algorithm of inverse iteration compounding is used, which contains the arithmetic mean and the geometric mean of the components of adjacent decomposition levels. The adopted order for inverse iteration compounding is from the last level to the first one. The weight of the arithmetic mean to the geometric mean is one to one. In this way, four compounding components can be obtained. Thirdly, a new frame image with higher motion contrast can be obtained by using 2D discrete static wavelet inverse transform of the four compounding components. Both flow phantom and live animal experiments are performed. The results show that classification error rate decreases by 83% and 71% respectively after component compounding in wavelet domain. Besides, the angiogram has an improved motion contrast and a better vessel connectivity, which may contribute to better and wider applications of OCT angiography. Furthermore, based on the developed system, the preliminary imaging studies on the model of local stroke are conducted. In this experiment, we record the 3D data of SD mouse brain before and after the local stroke and on the tenth day. As a consequence, a clear presentation for the whole process of stroke model formation, vessel damage and vessel recovery is achieved, which may be beneficial to studying the mechanism of local stroke model.