Single-Sample Face Recognition with Image Corruption and Misalignment via Sparse Illumination Transfer

Abstract

Single-sample face recognition is one of the most challenging problems in face recognition. We propose a novel face recognition algorithm to address this problem based on a sparse representation based classification (SRC) framework. The new algorithm is robust to image misalignment and pixel corruption, and is able to reduce required training images to one sample per class. To compensate the missing illumination information typically provided by multiple training images, a sparse illumination transfer (SIT) technique is introduced. The SIT algorithms seek additional illumination examples of face images from one or more additional subject classes, and form an illumination dictionary. By enforcing a sparse representation of the query image, the method can recover and transfer the pose and illumination information from the alignment stage to the recognition stage. Our extensive experiments have demonstrated that the new algorithms significantly outperform the existing algorithms in the single-sample regime and with less restrictions. In particular, the face alignment accuracy is comparable to that of the well-known Deformable SRC algorithm using multiple training images; and the face recognition accuracy exceeds those of the SRC and Extended SRC algorithms using hand labeled alignment initialization.

Semi-Supervised Sparse Representation Based Classification for Face Recognition With Insufficient Labeled Samples

“…The proposed method is flexible, in which the gallery dictionary learning method is complementary to existing methods which focus on learning the (linear) variation dictionary, such as ESRC [13], SVDL [14], SILT [15], [16], RADL [17] etc. For the ESRC, SVDL and RADL methods, we have shown by experiments that the combination of the proposed gallery dictionary learning and ESRC (namely S 3 RC), SVDL (namely S 3 RC-SVDL) and RADL (namely S 3 RC-RADL) achieve significantly improved performance on all the tasks.…”

“…Researchers can also use other alignment techniques (e.g. SILT [15], [16], DSRC [2] by aligning the gallery first, then align the query data to the well aligned gallery; or TIPCA [60]), but the comparison of different alignment methods is beyond the scope of this paper.…”

“…Motivated by ESRC, Yang et al proposed the Sparse Variation Dictionary Learning (SVDL) model to learn the variation dictionary V, more precisely [14]. In addition to modeling the variation dictionary by a linear illumination model, Zhuang et al [15], [16] also integrated auto-alignment into their method. Gao et al [26] extended the ESRC model by dividing the image samples into several patches for recognition.…”

“…6 Initialize each Gaussian of the GMM by N (p i , I) for i = 1, ...K, where p i is the i-th column of P. 7 Initialize the prior of GMM, π i = n i /n (for the insufficient training samples problem), or π i = 1/K (for SLSPP problem) for i = 1, ...K. 8 repeat 9 E-Step: Calculateẑ ij by Eqs. (14) and (15). dictionary, but it needs sufficient label/training data.…”

Sparse Illumination Learning and Transfer for Single-Sample Face Recognition with Image Corruption and Misalignment

“…To the best of our knowledge, the paper [40] was the first to propose a solution to perform small-sample-set facial alignment and recognition via a sparse illumination transfer. However, the construction of the illumination dictionary in [40] was largely ad hoc via a simple concatenation of the auxiliary illumination samples. It was suggested in [40] that a sparse illumination representation can be found to compensate for the missing illumination model in single gallery images.…”