A method and device for coding and decoding a coded data stream representative of at least one image split into blocks. For a current block, a group of pixel values in the block is determined from previously decoded pixels, and for each pixel, a prediction residue is decoded. A prediction value for the pixel is determined according to a first prediction mode by predicting the pixel from at least one other previously decoded pixel belonging to the current block. Information is decoded from the stream indicating whether the pixel is predicted according to a second prediction mode using a prediction resulting from the group of pixel values in the block. When the pixel is predicted according the second mode, the prediction value for the pixel is replaced with a selected value of the group. The pixel is reconstructed using the prediction value associated with the pixel and the prediction residue.