A method of processing components of an image for coding of an image portion of the image, the image being composed of a first component and a different second component. Samples of the second component are predictable from samples of the first component using a prediction model having a model parameter value. The method comprises: determining a plurality of subsets of samples usable for computing the model parameter value, wherein each subset of samples comprises samples of neighboring borders of the first component and the corresponding second component of the image portion, according to at least one predetermined sampling pattern and corresponding predetermined shifting offset, selecting, based on predetermined criteria, a subset of samples from among the plurality of subsets of samples; and predicting samples of the image portion using the prediction model and the model parameter value determined from the selected subset of samples.