A non-reference video quality evaluation algorithm for HD video (based on QP and number of skipped macro blocks)

This paper records a non-reference video quality evaluation algorithm. This is our own laboratory in the first two years of a sister-in-the-job, the algorithm is still relatively accurate, in this record.

Note the premise of this algorithm is high-definition video. And it's a coding method of H.

The method mainly uses the parameters of two code stream to evaluate the quality: Quantization factor (QP) and the number of skipped macro blocks (Skip_num).

Four sequences were used as test sequences: Spray,running,dunk,flower.

The following figure shows the relationship between QP and MOS (subjective evaluation of average opinion points). It can be seen that the MOS values under the same QP in different sequences are different because of the content. But the trend is similar: with the increase of QP, MOS decreases.

The following figure shows the relationship between Skip_num and MOS (subjective evaluation of average opinion points). It can be seen that the MOS values under the same skip_num of different sequences are different because of the content. But the trend is similar: as the skip_num increases, the MoS decreases.

The final algorithm formula is as follows:

The relationship between the predicted results and the subjective quality of the algorithm can be seen with good accuracy: