A method for improved prediction in boundary samples in prediction units (PUs) in pixel based prediction. The method includes computing predicted values for a plurality of border pixels in a PU using corresponding left, upper, and upper left adjacent neighboring pixels in the PU. The method further includes computing predicted values for a second plurality of pixels in the PU using sample based angular prediction (SAP). The second plurality of pixels includes all other pixels in the PU besides the plurality of border pixels.