a creature can indeed notice that it is in the presence of something we call red, according to our norms. but in order for us to conclude that said creature understands the concept "red", it must also demonstrate the ability to draw the same sort of inferences we can draw from and to red, like "therefore it is not green", etc.
I'm not sure he would agree that we must have the concept of red before we can perceive red things, or say that they are in the vicinity. maybe they are coeval, the mechanism involved remains a mystery, as far as I know. but I think his point is not about how this happens, only that it is a kind of structural condition of adequacy that must be satisfied by any attempt to explain such stuff. Something like "what conditions must be satisfied in order for us to conclude that creature understands what "red" means?"
McDowell covers this in great detail in Mind and World. Brandom does too in various places. They disagree somewhat ferociously, which is interesting since they're both on the World's Greatest Philosophy Department at Pitt.
FOLLOW UP Before I offer a sketch of an argument, let's do a little thought experimentation. Children routinely learn to utter sentences they do not understand. This includes observation report tokenings like "that's red". So let's assume Baby Jones is just learning to speak, and Mama Jones wants to teach him to say "that's red" in the presence of red things. She shows him a red triangle, says "that's red", and tries to get him to imitate her. She does this repeatedly, giving him positive and negative feedback. Let's stipulate further that she does not use words like "no" or "not red" - she only wants to teach "that's red".
Eventually he reaches the point where he always gets it right. Still we would not count his tokenings of "that's red" as expressing observational knowledge, because he still has no idea what "red" means; he just "knows" that uttering "that's red" in the presence of red triangles elicits approval. Note that this is an acquired skill or disposition, and that there are two (unexplained) mechanisms involved, differential sensing and the acquisition of reliable differential dispositions.
Now suppose Mama Jones shows him a blue triangle. He might be perplexed and say nothing, but let's assume he says "that's red". We, but not he, might take that as the expression of an implicit inference: the red triangles were triangular, that thing has the same shape, therefore "that's red" is likely to win approval. In any case he is not yet a reliable reporter of red things regardless of shape. So Mama disapproves, and proceeds to show him a variety of shapes and colors, approving only when he says "that's red" in the presence of red shapes. With enough experience (that is, sensings plus approval/disapproval) he acquires a reliable responsive disposition to say "that's red" only in the presence of the red shapes she has shown him. Still this does not count as knowledge, as I understand Sellars, because it does not involve inference. His red reports remain non-inferential reports.
It is only when he begins to grasp that his disposition to declare "that's red" is reliable that he begins to make the transition from sentience to sapience. That happens when he begins to reliably receive approbation whenever he says "that's red". He never or rarely gets it wrong, so he begins to take himself as a reliable red reporter. He begins to operate on the (implicit!) inferential rule "if I am disposed to say 'that thing is red', then that thing is in fact red." Now when he in fact feels himself so disposed, he says "that's red" (thus making a practical inference), and he is in a position to justify his claim by offering his reliable disposition as a reason. So at that point his tokenings of "that's red" express knowledge, based on his own authority as a reliable reporter plus his awareness of that reliability.
Notice that before he reaches this stage his tokenings do not count as moves in the game of giving and asking for reasons. If challenged at the earlier stage (where he only has a reliable disposition without realizing that such a disposition counts as justification), he would not even be able to grasp what a challenge is - he would take "justify your claim" as disapprobation rather than a request for justification. It is only when he grasps (no matter how implicitly, and no matter what the mechanism) that his reliable disposition serves as justification for saying "that's red" that he is even in a position to understand concepts like "challenge" and "justification".
So there is both a causal and a normative dimension to knowledge. One of Sellars' basic points is that while the causal dimension may be necessary in some sense it is not sufficient to explain knowledge. Observation reports like "that's red" are arrived at non-inferentially, but qua knowledge they nonetheless presuppose knowledge of other facts, and must be justifiable. Knowing that that's red presupposes additional knowledge of fact, in particular "facts of the form X is a reliable symptom of Y (section 36).
Note the critical role of the notion of ability or capacity or possibility. For a statement to count as knowledge it must be possible to offer some justification for it. The justification offered in any particular case could be wrong, but that's not germane; what matters is the challenge-justification structure of knowledge.
You wrote "For [Sellars], I can notice that I'm in presence of red colour only after I already have the concept 'red'." I hope the above helps show why that isn't quite right. You can certainly sense red things, and maybe even "notice" them, before you have the concept "red" - "...there is no reason to suppose that having the sensation of a red triangle is a cognitive or epistemic fact" (section 7) - but you can only know that they are red if you have acquired the concept.
You also wrote "[Sellars] claims that a sentence needs to have an authority in order to take part in the game, and therefore the concept 'red' comes before the red sensation." The concept comes before the sensation only in the order of explanation and justification. The sensation can temporally precede the concept, but can only be recognized as an instance of a concept after the concept has been acquired. That's the topic of section 36.
That's my take, anyway, heavily influenced by the study guide in Brandom's edition. Sellars is notoriously hard to read.