A main directional maximal difference analysis for spotting facial movements from long-term videos

A Survey of Automatic Facial Micro-Expression Analysis: Databases, Methods, and Challenges

“…In their later work (Polikovsky and Kameda, 2013), a tracking algorithm is applied to track the facial points that had been manually detected at the first frame throughout the whole sequence. To prevent the hassle of manually detecting the facial points, majority of the recent works (Davison et al, 2015, 2016a,b; Liong et al, 2015, 2016b,c; Wang et al, 2016a; Xia et al, 2016) opt to apply automatic facial landmark detection. Instead of running the detection for the whole sequence of facial images, the facial points are only detected at the first frame and fixed in the consecutive frames with the assumption that these points will only change minimally due to the subtleness of MEs.…”

“…Another popular facial segmentation method is splitting the face into a specific number ( m × n ) of blocks (Moilanen et al, 2014; Davison et al, 2015, 2016a; Wang et al, 2016a; Li et al, 2017). In the blocking representation, the motion changes in each block could by observed and analysis independently.…”

“…This work also indicated that LBP consistently outperforms HOOF in all the datasets with higher AUC (area-under-the-ROC-curve) values and lower false positive rates. To spot facial micro-expressions on the new CAS(ME) 2 database, the same spotting approach (Moilanen et al, 2014) is adopted by Wang et al (2016a). Using their proposed main directional optical flow (MDMD) approach, ME spotting performance on the CAS(ME) 2 is 0.32, 0.35, and 0.33 for recall, precision and F1-score, respectively.…”

“…Using their proposed main directional optical flow (MDMD) approach, ME spotting performance on the CAS(ME) 2 is 0.32, 0.35, and 0.33 for recall, precision and F1-score, respectively. For all these works (Moilanen et al, 2014; Wang et al, 2016a; Li et al, 2017; Qu et al, 2017), the threshold value for peak detection is set by taking the difference between the mean and max value of the contrasting difference vector and multiplying it by a fraction in the range of [0,1]. Finally, this value is added with the mean value of the contrasting difference vector to denote the threshold.…”

“…Besides, pre-defining the ME window intervals (which obtains the FD values) may not augur well with videos captured at different frame rates. To address the potentiality of a false peak, these works (Moilanen et al, 2014; Davison et al, 2015; Wang et al, 2016a; Li et al, 2017; Qu et al, 2017) proposed to compute the baseline threshold based on a neutral video sequence from each individual subject in the datasets.…”