The recognition of the book spine and call number is crucial for a robot working in library, which has become a hot spot of present research. In this paper, a simple projection operator is proposed and studied to improve the detection of the book spine and call number. A column vector of book is constructed on the basis of the binary image of books in bookshelf, then a project column vector is obtained through a column projection operator. The book spine can be recognized correctly by analyzing the curve of the project column vector. Similarly, the row vector and the column vector of the book spine are constructed on the basis of the binary image of the book spine with a call number. Then, a project row vector is obtained through a row projection operator. The row position information of the call number can be recognized by analyzing the curve of the row projection vector. In a similar way, a project column vector is obtained through a column projection operator, and the column position information can be recognized by analyzing the curve of the column vector. On the basis above, each image element of the call number can be cut and the matching degree between the image element and standard char template can be calculated after normalization, through which the char corresponding to the maximal match degree will be regarded as the recognition result. Experiments show that the projection operator can quickly and correctly recognize the book back as well as the call number.