Automorphisms of SO_n(k,f) Let $k$ be a field, $n\in\mathbb{N}$ and $f:k^n\times k^n\to k$ a non-degenerate symmetric bilinear form. Let
$$O_n(k,f):=\{ g\in GL_n(k) \mid \forall x,y\in k^n : f(x,y)=f(g.x,g.y) \}$$
and
$$SO_n(k,f):=O_n^+(k,f):=O_n(k,f) \cap SL_n(k)$$
be the associated (special) orthogonal group, i.e. linear transformations which preserve the form (with determinant 1). According to a paper by María J. Wonenburger ("The automorphisms of the group of rotations and its projective group corresponding to quadratic forms of any index", Canad. J. Math. 15 (1963), 302–303), building upon the works of Dieudonné and Rickart, the following is true:

Theorem: Suppose $\operatorname{char} k\neq 2$ and $n\geq 5$. Then for every $\phi\in \operatorname{Aut}(SO_n(k,f))$ there exists a homomorphism $\chi:SO_n(k,f)\to \{\pm I_n\}$ and a semi-similitude $T$ of $f$ such that $$\phi(g)=\chi(g)\ TgT^{-1}.$$

Here, $I_n$ is the identity matrix, and a semi-similitude of $k^n$ is a permutation of $K^n$ for which there are $\sigma\in \operatorname{Aut}(k)$ and $\lambda \in k\setminus\{0\}$ such that for all $x,y\in k^n$ one has $$f(Tx,Ty) = \lambda\ f(x,y)^\sigma.$$
If we call automorphisms that can be written as in the theorem ``standard'', then the theorem shows that for $n\geq 5$, all automorphisms are standard automorphisms.

Question: What is known when $n=3$ and $n=4$? Are there any ``non-standard'' automorphisms? If yes, when do they exist and what do they look like?

Both (counter)examples as well as partial affirmative answers (e.g. with restrictions on the Witt index or on the the field) are of interest.
EDIT: removed nonsensical statement about algebraic groups and types
UPDATE: I just discovered a paper by Li Zunxian, "Quaternion algebra and automorphisms of $\rm PO^+_4(V),\;\rm PO'_4(V)$ and $\rm P\Omega_4(V)$" (direct link), which seems to construct ``exceptional'' automorphisms of some projective orthogonal groups for $n=4$. I have not yet had time to study it in detail or figure out if this lifts to the non-projective case.
 A: This is at least a partial answer.  There are two distinct viewpoints here: the concrete one involving forms and automorphism groups (which came first historically and usually requires characteristic $\ne 2$ to avoid tricky points) and the much more general one involving simple algebraic groups over a field $k$ (where Borel and Tits covered a great deal of territory when the groups are $k$-isotropic).    But the cases $n=3,4$ you raise are degenerate from either viewpoint.   When $n=3$ you don't get Lie type $B_2$ (that corresponds to $n=5$) but rather $B_1 = A_1$ along with a line on which nothing happens.   When $n=4$ what you get is Lie type $D_2 = A_1 \times A_1$, also a degenerate case.   In these low dimensions the automorphism group problem comes down to a form of $\mathrm{SL}_2$ over the field.  
For convenience, it might be useful to link directly to the online version of Wonenburger's short paper here.
ADDED: Though it gets far away from the theme of groups preserving bilinear forms, there is a thorough treatment of the rank 1 degenerate cases (in concrete style) by O.T. O'Meara, Lectures on linear groups. Expository Lectures from the CBMS Regional Conference held at Arizona State University, Tempe, Ariz., March 26–30, 1973. Conference Board of the Mathematical Sciences Regional Conference Series in Mathematics, No. 22. American Mathematical Society, Providence, R.I., 1974.
See especially his section 5.6, along with the comments and references in 5.8.
The nature of the field $k$ inevitably comes into play here.   As in other sources he mentions, O'Meara discusses more generally the isomorphisms between various linear groups and not just the automorphisms of a single group.
