Local Evidence Aggregation for Regression-Based Facial Point Detection

Abstract

Abstract-We propose a new algorithm to detect facial points in frontal and near-frontal face images. It combines a regressionbased approach with a probabilistic graphical model-based face shape model, that restricts the search to anthropomorphically consistent regions. While most regression-based approaches perform a sequential approximation of the target location, our algorithm detects the target location by aggregating the estimates obtained from stochastically selected local appearance information into a single robust prediction. The underlying assumption is that by aggregating the different estimates, their errors will cancel out as long as the regressor inputs are uncorrelated. Once this new perspective is adopted, the problem is reformulated as how to optimally select the test locations over which the regressors are evaluated. We propose to extend the regressionbased model to provide a quality measure of each prediction, and use the shape model to restrict and correct the sampling region. Our approach combines the low computational cost typical of regression-based approaches with the robustness of exhaustive-search approaches. The proposed algorithm was tested on over 7,500 images from 5 databases. Results showed significant improvement over the current state of the art.

Multi-Kernel Appearance Model

“…As pointed by [5], it is important to note that the curves of the works of [27] and [2] are hardly comparable as they re-annotated some of the data. Fig.…”

“…This is a particularly challenging dataset as it is captured under cluttered backgrounds and various real-world illumination environments. We thus compare our system with several recent methods: the original CLM from [11], LEAR from [27], BoRMaN from [46], STASM from [30] and the system of [2] and [14]. For a fair comparison, we use the error measure for only 17 points, as [11] explains.…”

“…Some methods directly model the positions of each facial point based on the others [43]. Other methods model these relations using graph-based methods [46,15,27,54]. The spatial restriction can also be included in a transparent way by always allowing the point configuration to be a linear combination of the learned shape models [3].…”

“…Some methods independently detect each point and apply spatial restrictions on their locations using graph-based methods [15,46,27,54]. In a recent work, [37] first detects each landmark using multi-scale gray-level features and multi-kernel Support Vector Machines.…”

Robust Face Alignment with Random Forest: Analysis of Initialization, Landmarks Regression, and Shape Regularization Methods

“…We first report the comparison results on the BioID Dataset [9]. Figure 13 shows the face alignment results of LFCR, Stasm [2], FaceTracker [4], BoRMaN [32], and LEAR [33]. From the figure, we can observe that LFCR performs the best, followed by Stasm, FaceTracker, LEAR, and BoRMaN.…”