Center Settled Multiple-Coil Spring Model to Improve Facial Recognition Under Various Complexities

A facial extracted image does not have the equalized distribution of features over the complete image. Instead, most striking features are located within the core of the facial part. As the distance increases from the core part, the strength of these features faded and its impact on the recognition model reduces. In this paper, a coil spring structured model is presented to generate the selective features based on structured weights. These weights are assigned under the pressure, position, direction and coverage parameters of magnetic coils. The magnetic coil effect is applied to extract the facial features. These features are collected and mapped with dataset images with region consideration. This mapping is done for the individual region with physical features and coil-spring based evaluation. As the method is center settled, so that the effective recognition rate is achieved missing facial information or the wrong captured images. The experimentation is applied to the complete facial image sets as well as improper, occluded and irregular captured facial images. The comparative analysis is provided on Aberdeen, Stirling, Iranian, ORL, FERET and LFW databases. The proportionate observations are taken against six different algorithms, including LDA, PCA, ICA, LDA–PCA, SVM and PNN classifiers. Multiple sample sets are considered over each dataset under distinctive variation aspects. These variations include expression, pose, illumination, occlusion, etc. The analytical evaluation is also taken for CNN and landmark based methods. The extensive experimentation shows that model has improved the accuracy and robustness up to an extent. The recognition rate for each variation aspect is improved.