The protection of well-designed trademarks is an essential method to promote the brand effect of enterprises on their products or services. However, in the development of market economy in China, there is a lack of perfect legal protection for trademark design, which leads many illegal businesses to counterfeit or imitate trademarks of other enterprises, thus causing certain economic losses to lawfully operated enterprises. In the field of image recognition, trademark recognition is different from the traditional barcode or QR code recognition, which is to obtain the specific information of goods by directly recognizing trademark graphics. This paper presents a theoretical approach of the image retrieval process using digital image processing technology. Firstly, the image is preprocessed, and the foreground object of the image is extracted by multi-threshold color segmentation method. Then, the contour features of the object are obtained by contour tracking method, and the contour is described by Fourier descriptor. The simplified BP neural network (BPNN) is used as a trademark recognition method to test on the recognition system. The test results show that the whole recognition system has less computation, fast recognition speed and high recognition rate of trademarks.