- This project aims to build a model that classifies the subjects into female-male and old-young from the images of the subjects' dorsal hand veins.
- The project can be summarized into 3 main steps:
- Dataset: Hand vein images of 200 subjects (104 male & 96 female// 61 old & 149 young)
- Preprocessing: Some processing (filters, extracting ROI) and augmentation techniques
- are applied on the images to enhance their quality.
- Deeplearning: A fine-tuned model that is used to classify the images that is based on VGG-16.
- The Dataset directory has several sub-directories:
- Enhanced_Augmented_roi dataset: the enhanced images for age and for gender classification.
- Original_Dataset: the original dataset as raw images without preprocessing/division/excluding_prediction_subjects.
- Prediction_subjects: images of 4 subjects that were cut from each category for prediction.
- Raw data (prediction excluded): Raw images divided into gender-age with excluding the prediction images.
Note: The number of the dataset and the reserved items for prediction can be identified using the Age_Gender.xlsx & ages-Gender-Data_modified.txt files.
- Firstly, The left-hand images were flipped using the Flipping_Left file.
- Secondly, for the enhancement track we input the flipped images to the enhancement/filters code(median, bilateral, CLAHE) that can be found in the Enhance_2020 file.
- Thirdly, we augment (Rotate, Translate, and Scale) the enhanced images using the augmentation code that can be found in the Augmentation_2020 file.
- Lastly, we take the region of interest "ROI" to the augmented enhanced images using the ROI_2020 file.
- In case of working with lbp images, then you'll have to use the lbp file, after the second step.
- Two subdirectories can be found: Age and Gender
- Both contain subdirectories for their "codes" and "results"
- You first need to train the model on the images you're using (either raw images, enhanced or lbp ones) using the train file.
- Then, you'll need to save the model at the desired epoch
- (depending on the accuracy and loss graphs obtained from training)
- And lastly, you can use this ".h5" saved model for prediction. During the prediction, a ".csv" file is made that comprises the image name, the prediction to it and its true label, a confusion matrix can be found at the end of this ".csv" file to better assess the prediction.