1. What are the two categories of hand recognition methods?
The two categories of hand recognition methods are direct recognition of processed images and analysis of processed images for feature values. Direct recognition is not suitable for object rotation or scaling and requires a large amount of calculation for identification. The second type analyzes processed images, selects feature values, and proceeds to identification. This method is suitable for identifying objects of different angles and sizes. However, complex images may require more extracted feature values, leading to longer identification times. To reduce identification time and improve accuracy, classifying captured images is considered. This approach reduces the number of samples during identification and eliminates redundant feature data. Various methods have been proposed for hand recognition, including neural network algorithms, B-spline curve methods, implicit polynomial and geometric features, and HMM algorithms. Hand shape recognition is crucial for identification, and features such as inverse of compactness are used for matching. Fuzzy Classifier algorithms and eigenvalue classification methods are also employed for hand recognition. These methods aim to achieve translation-invariant, rotation-invariant, and scale-invariant results, ensuring TRS-invariant hand recognition.
read more
2. What is the main method used for feature classification in the research?
The main method used for feature classification in the research is the fuzzy k-neighbor cluster analysis method (FKNN). It is mainly used to classify feature value clusters in advance to simplify the complexity of the identification process. FKNN is a commonly used classification method that provides a simple step by dividing highly correlated samples into a group based on the degree of correlation of each sample in the feature vector space. It allows for the description of data points in the middle zone through the processing of membership degree and can describe more data points compared to traditional hard clustering. The classify cost function is defined using a cost function suitable for one-dimensional or higher dimensional features, and the classify cost is represented as a fuzzy number. The fuzzy edit costs are defined as fuzzy numbers, and the fuzzy shortest paths can be found by ranking fuzzy numbers using the integral values of the inverse of membership functions.
read more
3. How many test images were used for evaluation in the experiment?
A total of 6,400 test images were used for evaluation in the experiment. This is achieved by using 100 hand images, each with 64 test images (16 different orientations and 4 different scales). The experiment aims to evaluate the performance of the proposed method in hand recognition tasks. The test images are generated by rotating the hands in different orientations and downscaling the images to 90%, 80%, and 70% of the original dimensions. The edge detection algorithm is used to find a rough sketch of the hand, followed by a dominant point algorithm to identify important points. The inverse of compactness is then calculated for each finger and palm, serving as a feature in the fuzzy Classifier algorithm. The fuzzy k-neighbourhood cluster analysis method is applied to classify the features, with the K value set as the number of feature vectors. The recognition rate is calculated, and experimental results show a correct recognition rate of about 94.3%. However, overlapping fingers were identified as the primary cause of recognition errors, leading to hand shape distortion.
read more
4. What is the role of hand recognition in human biometrics?
Hand recognition plays a crucial role in human biometrics as it helps reduce the complexity of the human body recognition problem. An effective hand recognition algorithm is essential for accurate identification and authentication. By utilizing advanced techniques like fuzzy classifiers and innovative features such as the inverse of compactness, hand recognition systems can effectively recognize hand shapes and improve overall security measures. The ability to classify input hand shapes with minimal discrepancy from reference shapes enhances the reliability and efficiency of biometric systems. Additionally, the proposed method's advantage of not requiring any parameters further simplifies the hand recognition process, making it more accessible and adaptable for various applications.
read more