Though prime ministers do not normally respond in person to a no-confidence motion, Mr. Barak had to attend today's session and listen to the scorn heaped on him because Limor Livnat of Likud had taken the unusual step of getting 40 members' signatures requiring Mr. Barak's presence.
The tour coincided with speculation about a possible Beatles reunion, as well as an agreement regarding the official dissolution of their partnership, which would be ratified following the receipt of all four members' signatures in December 1974.
In accordance with the hyperspectral imaging terminology, we shall interchangeably refer to these end-member profiles as end-member signatures.
The problem of estimating end-member signatures is then reduced to finding the data vectors that form the vertices of the simplex.
The aforementioned normalization technique can be understood intuitively by considering a simple case where we have a mixture of 2 end-members (K = 2), for which the end-member signatures s1 and s2 are represented as vectors OS1 and OS2 in Fig. 1.
However, in cases where there is no prior knowledge about the possible number of end-members, any or more than one of the above techniques can be used in conjugation with the interpretation of the end-member signatures obtained by NMF to decide on an appropriate number of end-members.
The first category includes algorithms that are based on the assumption that the set containing the end-member signatures 𝔖 ≡ { s k ∈ ℝ T, k = 1, 2,..., K} is a subset of the set 𝔛 ≡ { x n ∈ ℝ T, n = 1, 2,..., N} containing the data vectors.
This effect is commonly referred to as signature variability in hyperspectral imaging literature and is characterized by what is called spectral shape invariance, which means that although the shapes of the end-member signatures do not vary much across different pixels, their scaling or amplitude vary significantly over the imaged area [ 16].
We shall call: S ≡ [ s1, s2, ⋯, s K] the mixing matrix, which holds the K end-member signatures in its columns, and A ∈ ℝ K× N shall be referred to as the abundance matrix, which contains the contributions of K end-members (along rows) for each of the N pixels (along columns).
By similar reasoning, the normalized end-member signatures, s̃1 and s̃2 (denoted by vectors OS′1 and OS′2) also lie on the unit L1-norm circle, whence it follows that any vector expressed as a non-negative linear combination of s1 and s2 (or lying in the blue shaded region) upon normalization will fall on the line joining S′1 and S′2, which represents a 1-D simplex.
