More Efficient Feature Method Than Haar-Feature For Face Detection

The seminal work by Viola-Jones had the most impact in face detection. It used Haar-Feature, which is fast to compute with the help of integral image method.

Is there a feature method that is more efficient (computationally) than Haar-Feature?

For example, for a $24 \times 24$ window, the computational complexity is more than $180,000$.

Is there any feature method which has the same efficiency (or better) as obtained by Viola-Jones but less expensive computational complexity-wise?

• LBP features are integer in contrast to Haar features, so both training and detection with LBP are several times faster then with Haar features. – Mike SQ May 3 '17 at 11:15
• What do you mean by "same efficiency but less expensive"? What distinction are you trying to draw between those two? And given your comment, have you answered your own question? If so, would you write an answer to your own question? – D.W. May 3 '17 at 14:10
• @D.W. for example, if u come up with a feature system that takes number of operations (computational-complexity) less than $180000$ keeping the detection rate (true-positive) same as Viola-Jones, we would consider that "same efficiency but less expensive". The comment is just a note , its not an answer obviously , I am searching, and gathering . – Mike SQ May 3 '17 at 14:54
• Ahh, ok. Detection rate would normally be called accuracy or detection rate or effectiveness, not efficiency; efficiency normally refers to the running time (computational complexity). Thanks. – D.W. May 3 '17 at 16:05
• The method by Viola and Jones gives pretty good results for face detection in a quite short time. I doubt that you can achieve significantly better results in the same time on the same hardware with other approaches. I think recent developments focus rather on the "pretty good results" part and use much better hardware and a bit more time. – Martin Thoma May 4 '17 at 7:23