This paper presents, for I-frames, a generic relationship among the quantization value, data bits, and their spatial complexity. This relationship provides a mechanism to select the quantization value for an I-frame (with different spatial complexity) at given bit budget before actually encoding it. Therefore it helps to adaptively distribute the bit budget among I-, P- and B-frames in a sequence. Experiments show that the adaptive intra frame quantization algorithm reduces the number of frame skippings significantly with comparable average PSNRs, and thus improves the overall subjective quality of the reconstructed video.