Completed local structure patterns on three orthogonal planes for dynamic texture recognition

Abstract

Abstract-Dynamic texture (DT) is a challenging problem in computer vision because of the chaotic motion of textures. We address in this paper a new dynamic texture operator by considering local structure patterns (LSP) and completed local binary patterns (CLBP) for static images in three orthogonal planes to capture spatial-temporal texture structures. Since the typical operator of local binary patterns (LBP), which uses center pixel for thresholding, has some limitations such as sensitivity to noise and near uniform regions, the proposed approach can deal with these drawbacks by using global and local texture information for adaptive thresholding and CLBP for exploiting complementary texture information in three orthogonal planes. Evaluations on different datasets of dynamic textures (UCLA, DynTex, DynTex++) show that our proposal significantly outperforms recent results in the state-of-the-art approaches.

Directional Beams of Dense Trajectories for Dynamic Texture Recognition

“…An another variant, called LBP-TOP [13], has been also presented to overcome the curse of dimensionality of VLBP by addressing LBP on three orthogonal planes. Various extensions based on two above works have been then proposed to advance the discriminative power: CVLBC [21], CVLBP [15], CLSP-TOP [14], HLBP [16].…”

“…k-nearest neighbors (k-NN): To be comparable with results of existing methods [14,15,16], we employ the simple of k-nearest neighbors, i.e. k = 1 (1-NN), in which chi-square (χ 2 ) measure as dissimilarity measure.…”

“…-50-class: Original 50 classes are utilized using two experimental protocols: leave-one-out (50-LOO) [4,17,29] and 4-fold cross validation (50-4fold) [14,16]. -9-class and 8-class: Original 50 classes of sequences are divided into 9 semantic categories [29,6] consisting of "boiling water" (8), "fire" (8), "flowers" (12), "fountains" (20), "plants" (108), "sea" (12), "smoke" (4), "water" (12), and "waterfall" (16), where the numbers in parentheses take account of sequences in each class.…”

“…The "plants" category is eliminated from 9-class to form more challenging 8-class scheme [29,6]. For these two schemes, following [29,30,14], a half of DTs is randomly selected for training and the remaining for testing work. The final evaluation of recognition is estimated by the average of rates in 20 runtimes.…”

“…Sixth, local-feature-based methods have been also considered with different LBP-based variants owing to their simplicity and efficiency since Zhao et al [13] proposed two LBP-based variants for DT depiction: Volume LBP (VLBP) and LBP on three orthogonal planes (LBP-TOP). Lately, several efforts based on extensions of these typical operators are addressed to enhance the discriminative power of DT description [14,15,16,17,18]. This paper addresses a new efficient framework using directional beams of dense trajectories for DT representation, in which local feature patterns of motion points are encoded along their trajectories in conjunction with directional features of their neighbors structured by the proposed LVP f ull -TOP operator with full-direction on three orthogonal planes of sequences.…”