To analyze the performance of the algorithm Extended Cohn-Kanade expression dataset\cite{Lucey_2010,Tian_2011} was used initially. Data set had only 486 sequences with 97 posers and causing accuracy just to reach up to 45% maximum. To overcome this problem of low accuracy, multiple datasets were downloaded from the internet\cite{Widener_2010,Kring_1991,Russell_2011,Schlosberg_1960,2017a,Simon_2005} and also authors own pictures at different expressions were included. As the number of images in dataset increase, the accuracy also increased. We kept 70% of 10K dataset images as training and 30 % dataset images as testing images. In all 25 iterations were carried out, with the different set of 70% training data and then error bar was computed as standard deviation. Figure 4 (a) shows optimization of the number of layers for CNN. For simplicity, we kept the number of layers and number of filters for background removal CNN and face feature extraction CNN to be the same. In this study, we varied the number of layers from 1 to 8. We found out that maximum accuracy was obtained around 4. It was not very intuitive, as we assume the number of layers is directly proportional to accuracy and inversely proportional to execution time. Hence we selected the number of layers to be 4. The execution time was increasing with the number of layers, and it was not adding great value to our study hence not reported in the current manuscript. Figure 4 (b) shows the number of filters optimization for both layers. Again 1 to 8 filters were tried for each of four-layer CNN networks. We found that four filters were giving good accuracy. Hence FERC was designed with four layers and four filters.  
As a future scope of this study, researchers can try varying number of layers for both CNN independently. Also, the vast amount of work can be done, if each layer is fed with a different number of filters. This could be automated using servers. Due to computational power limitation of the author, we did not carry out this study, but it will be highly appriciated if other researchers to come out with a better number than 4 (layers), 4(filters) and increase the accuracy beyond 96%, which we could achieve. 
Figure 4 (c and e) were normal front facing cases with angry and surprised emotions and the algorithm could easily detect them (fig 4 d and f ). The only challenging part in these images, was skin tone detection, because of the grayscale nature of these images. With color images, background removal with the help of skin tone detection was really easy, but with grayscale images we observed false face detection in many cases. Image, such as, figure 4 g was challenging because of the orientation. Fortunately, with 24 dimension EV feature vector, we could correctly classify 30 degree oriented faces using FERC. 
We do accept the method has some limitations such as high computing power during CNN tuning and also, facial hair causes a lot of problems. But other than these problems the accuracy of our algorithm is very high (i.e. 96% ) which is comparable to most of reported literature\cite{Mal_2017,2010,Martinez_2016,Saha_2013,OuYang_2013,Dantes_2017}. One of the major limitations of this method is when all 24 features in EV vector is not obtained due to orientation or shadow on the face. Authors are trying to overcome shadow limitation by automated gamma correction on images (manuscript under preparation). For orientation, we could not find any strong solution, other than assuming facial symmetry. Due to facial symmetry we are generating missing feature parameters by copying the same 12 values for missing entries in the EV matrix.(e.g. Distance between left eye to left ear (LY-LE) is assumes same as right eye to right ear(RY-RE) etc.) Algorithm also failed, when mutiple faces were present in same image, with equal distance from camera.

Conclusions

FERC is a novel way of facial emotion detection that uses advantages of CNN and supervised learning (feasible due to big data). The main advantage of the FERC algorithm is that, it works with different orientations (less than 30 degrees) due to unique 24 dimensions EV feature matrix. The background removal added a great advantage, in accurately determining the emotions. FERC could be starting step for many of the emotion-based applications such as lie detector and also mood based study for students etc.

Acknowledgements

Author would like to thank Dr. Madhura Mehendale for her constant support on database generation and corresponding ground truths cross validation.