Classification module

Classification module architecture is depicted on figure A.5 on page . The interface defines:

• $I_k$ = classify( ${I_{test}}$, ${I_{train_1}}$, ${I_{train_2}}$, ...)
• $I_k$ = classify( $\mathbf{\mu_{test}}$, $\mathbf{\mu_{train_1}}$, $\mathbf{\mu_{train_2}}$, ...)
• $I_k$ = classify( $[e_{ij}]$ )
• $\{(I_1, p_1), (I_2, p_2), \dots\}$ = classify( ${I_{test}}$, ${I_{train_1}}$, ${I_{train_2}}$, ...)
• $\{(I_1, p_1), (I_2, p_2), \dots\}$ = classify( $\mathbf{\mu_{test}}$, $\mathbf{\mu_{train_1}}$, $\mathbf{\mu_{train_2}}$, ...)
• $\{(I_1, p_1), (I_2, p_2), \dots\}$ = classify( $[e_{ij}]$ )

where $I_k$ is the image from training set, $p_k$ is the classification confidence, $I_{test}$ is the image to classify $\mathbf{\mu_{test}}$ is the test car feature vector and $[e_{ij}]$ is the error matrix. The latter case of classification methods reflects the fact that some of classification methods are able to return more than one solution and tell us the classification probability.

Figure A.5: Classification module