Owing to the sparsity of the space distribution of point scatterers, compressed sensing (CS) method is successfully applied in inverse synthetic aperture radar (ISAR) imaging. However, in addition to sparsity, ISAR images usually exhibit group sparse structure. Here, the authors propose a novel two-dimensional (2D) group primal dual active set with continuation (2DGPDASC) algorithm to recover an ISAR image, which always exhibit 2D group sparse structure. This algorithm is based on the regularised least-squares problem with an ${\rm \ell }^0\left({{\rm \ell }^2} \right)$ℓ0ℓ2 penalty model. At each iteration of the proposed method, it involves solving a least-squares problem on the active set only, and exhibits a fast local convergence within a finite step. Experimental results validate the effectiveness and superiority of the proposed method.