A method for decoding a picture based on a cross-component linear model (CCLM) mode includes deriving neighboring luma reference samples of a luma block, deriving down-sampled neighboring luma reference samples, deriving a linear model parameter based on the down-sampled neighboring luma reference samples and the neighboring chroma reference samples, where the neighboring luma reference samples includes top neighboring luma reference samples, and left neighboring luma reference samples, and where when the top boundary of the luma block overlaps with a boundary of a coding tree unit (CTU), the number of the top neighboring luma reference samples used for deriving the down-sampled neighboring luma reference samples among the neighboring luma reference samples is less than that of the left neighboring luma reference samples used for deriving the down-sampled neighboring luma reference samples.