The question is about stationary point of a least squares being identical to the stationary point of an error function as stated in: Computer Vision: Algorithms and Applications by Szelinski et al. p.318

The premise is:

For a cost function with robust error function $h(e_i)$ $$E(\delta p)=\sum\limits_{i=0}^nh(\|r\|)$$

Where h is a robust penalty function and r is the residual. $$w(x) = \frac{h'(x)}{x}$$


$$ \frac{\delta h(\|r\|)}{\delta p} = w(\|r\|)\|r\| \equiv \frac{\delta E_w}{\delta p}$$

This is what is stated in the text. However, when I try to derive this I end up with:

$$w(\|r\|)\|r\| + h''(\|r\|)\|r\|$$ due to the product rule.

So in this case, is the second derivative just ignored?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.