Dilated deep learning approach for segmenting and classifying the facial feature

Task of automatically detecting facial features from facial images is a remarkably laborious and formidable endeavor, requiring a great deal of attention to detail and diligence. In this paper, propose a dilated Unet, combined with Bi-LSTM, to achieve unparalleled accuracy in predicting facial features from facial images. The kernel layer 3 × 3 and 5 × 5 layers are concatenated with the subsequent layer by dilated U-Net architecture. The dataset utilized in the paper comprises a total of 13,233 target face images, along with an additional 5749 images of various individuals. The present study identified 1680 images that exhibit similarity in the database, which have been taken into account for our analysis. In preparation for training on our dataset, undertook pre-processing and image augmentation to optimize their quality and suitability. The dilated Unet algorithm enabled us to effectively segment the features of each face image, which we then subjected to classification by the Bi-LSTM algorithm. The effectiveness of the framework was assessed by employing a variety of performance signs, including precision, F1-scoring recall, exactness, and accuracy. Hence, a comparison evaluation with other contemporary strategies to verify the findings.