An image decoding method, according to the present document, comprises a step of generating residual samples for a current block on the basis of residual information, wherein the residual samples are generated on the basis of information on a transform kernel set and transform coefficients for the current block, the transform coefficients are derived on the basis of the residual information, and the information on the transform kernel set represents a transform kernel set to be applied to the current block from among transform kernel set candidates. At least one bin from among bins of a bin string of the information on the transform kernel set is derived on the basis of context coding, and the context coding is performed on the basis of at least one context model.