The invention belongs to the technical field of image saliency detection, and relates to an image saliency detection method based on semi-supervised. The method comprises the following steps: S1, establishing a coordinate system: establishing a plane coordinate system, establishing a plane coordinate system based on an operating system, establishing a plane coordinate system abscissa X, a longitudinal axis Y, and a coordinate origin O; S2, inserting an object and dividing an area; S3, object processing in the area unit; S4: prominent point highlights, background grammaticalization; 5, saving the area modular combination. In this scheme, the detection object is operated separately through the division of the region blocks, and the region blocks can be enlarged, each region block is taken asan individual unit, so that each region block does not affect each other, the boundary treatment is relatively completed, and the region blocks are operated separately, so as to reduce the recognition error of the identification object.