VIDEO                                                                  

Frame Organization

In the H.261 standard, video frames are processed by first dividing the image into smaller sections or “blocks”. A block is defined as an area with dimension 8 x 8 in pixels.

A CIF picture can then be defined as having dimensions 44 x 36 in blocks. Blocks belong to a larger group known as macroblocks. A macroblock consists of six blocks, 4 blocks for the Y luminance information, 1 for the Cr chrominance information, and 1 for the Cb chrominance information. These six blocks combine to form a 16 x 16 pixel matrix formed from the Y information with the Cr and Cb information averaged out over the larger block. Thirty-three macroblocks form a larger block known as a GOB or Group of Blocks.  A CIF picture would have a dimension of 6 x 2 in GOBs.

H.261 compression and encoding

The H.261 compression algorithm is DCT-based and resembles MPEG to some degree. DCT stands for Discrete Cosine Transformation. It is a popular method because it processes data in such a way that it can be more efficiently compressed using run length encoding (RLE). DCT transforms a block of pixel intensities into a block of frequency transform coefficients. The transform is then applied to new blocks until the entire image is transformed. Huffman/RLE encoding can then be performed on the processed data.

            A major difference between H.261 and MPEG is that the quantization value used is variable and is determined by the amount of data reduction required to fit the available video bandwidth. H261 has rate control which allows it to cope with a variable video bandwidth. H.261 trades picture quality against motion, which results in moving pictures having poor image quality as compared to still pictures. 

            The H.261 encoding process uses past frames to encode differences, much like the way MPEG does it. Unlike MPEG, H.261 only has two types of frames. These are the intra-coded frames and the inter-coded frames. MPEG utilizes three frames, the I or intra-frame, the P or predicted frame, and the B or bi-directional frame. H.261 only uses past frames as references for its motion estimation algorithm. MPEG’s B-frame uses the nearest preceding I or P frame and the next future I or P frame as its reference. H.261’s intra-coded frames are frames which are wholly encoded with no reference. Inter-coded frames are frames whose encoding is based on the previous frame. The H.261 standard states that each macroblock must be intra-coded at least every 132 frames to prevent errors from accumulating. Terminals may also ask for a complete picture update, wherein an intra-coded frame is sent.

 

Figure 3. H.261 encoder

          If a frame is being encoded as an intra-frame, the macroblocks go through a DCT transformation, quantization using a value determined by the rate control feedback and finally through the zigzag Huffman encoding to produce the final bitstream output. This output is replicated and sent to a decoder so it can be utilized by the inter-encoder.

Figure 4. H.261 decoder

          The H.261 decoder is much simpler. First of all, the encoder is not merely an encoder but also has a built-in decoder. This built-in decoder is used for decoding the bitstream output to recreate the reference frames needed for motion compensation. A frame can take one of two paths when it reaches the decoder. This all depends on the type of frame. Inter and Intra coded frames take different paths during decoding. The intra-coded macroblock is decoded by simply reversing the DCT-quantization-encode process. The decoded data is then used  to build up the frame. This frame can then be displayed. A copy of this frame is also stored for use as reference when decoding inter-coded macroblocks and frames.