The map \[ \psi: \PP^m \times \PP^n \to \PP^{(m+1)(n+1)-1}, \qquad ([x_i],[y_j]) \mapsto [x_i y_j] \] is well defined and injective, and its image is the closed set cut out by the \(2 \times 2\) minors \[ z_{ij} z_{kl} - z_{il} z_{kj} = 0 . \] A product of projective varieties is therefore projective.
For \(m = n = 1\) there is one minor, and the image is the quadric surface \[ z_{00} z_{11} - z_{01} z_{10} = 0 \] in \(\PP^3\). The two rulings of that quadric are the images of \(\ts{p} \times \PP^1\) and \(\PP^1 \times \ts{q}\), and they make it a surface whose Picard group is \(\ZZ^2\) rather than \(\ZZ\).
The minors say the matrix \((z_{ij})\) has rank one, which is the coordinate-free statement: the Segre image is the locus of rank-one tensors in \(\PP(V \tensor W)\).
Dependencies
D-CP2MHProjective varieties and homogeneous idealsuses