The invention provides a diagonalizable LDPC (Low Density Parity Check) observation matrix generating method based on compressed sensing. The method comprises the following steps: firstly, inputting an image to be processed, and determining the size of an observation matrix; secondly, obtaining LDPC sub-matrixes according to an LDPC test matrix generating method; and lastly, placing the obtained sub-matrixes on diagonals of the observation matrix in sequence, and zeroing remaining positions to obtain a final observation matrix. Quantitative and qualitative analyses are performed on a large number of images based on a block compressed sensing theory, so that the computing amount and storage amount of the observation matrix can be reduced greatly. Compared with other existing observations, the observation matrix has a good reconstruction effect. Meanwhile, the constructed matrix is highly sparse; elements are all binary; and the matrix is simple.