Blur Filter Research Articles

Integration of auditory and visual cues in spatial navigation under normal and impaired viewing conditions.

Auditory landmarks can contribute to spatial updating during navigation with vision. Whereas large inter-individual differences have been identified in how navigators combine auditory and visual landmarks, it is still unclear under what circumstances audition is used. Further, whether or not individuals optimally combine auditory cues with visual cues to decrease the amount of perceptual uncertainty, or variability, has not been well-documented. Here, we test audiovisual integration during spatial updating in a virtual navigation task. In Experiment 1, 24 individuals with normal sensory acuity completed a triangular homing task with either visual landmarks, auditory landmarks, or both. In addition, participants experienced a fourth condition with a covert spatial conflict where auditory landmarks were rotated relative to visual landmarks. Participants generally relied more on visual landmarks than auditory landmarks and were no more accurate with multisensory cues than with vision alone. In Experiment 2, a new group of 24 individuals completed the same task, but with simulated low vision in the form of a blur filter to increase visual uncertainty. Again, participants relied more on visual landmarks than auditory ones and no multisensory benefit occurred. Participants navigating with blur did not rely more on their hearing compared with the group that navigated with normal vision. These results support previous research showing that one sensory modality at a time may be sufficient for spatial updating, even under impaired viewing conditions. Future research could investigate task- and participant-specific factors that lead to different strategies of multisensory cue combination with auditory and visual cues.

A NEW METHOD FOR MEDICAL IMAGE FUSION BASED ON GAUSSIAN BLUR FILTER AND ROBINSON COMPASS OPERATOR

Medical image fusion is a process of extracting features from multi-modal medical images and combining them into a composite image. It brings huge support in medical imaging and clinical diagnosis. However, the extraction of both structural and functional information from input MRI and PET images using multi-scale transform fusion methods poses a challenge of providing high-quality decomposition layers since during the decomposition process, images can still lose information such as blur or noise at the edges of the image. To address this limitation, we present a new method to improve the visual information fidelity of medical image fusion. Firstly, the YCbCr color space is utilized to prevent distortion when merging color and grey images. The second algorithm uses the CLAHE model, which allows the input images to have good contrast. Then, a Gaussian blur filter is employed to decompose the images into base and detail layers. The use of Gaussian blur ensures a smoothing filter of the edges. After that, the Robinson compass operator is applied to create the fusion rule of detail components. Finally, the fused base and detail layers are concatenated together to form the final composite image. The experimental results show that the proposed approach outperforms the latest methods in bringing visual information fidelity of the input images to the fused image, which is helpful in supporting doctors and radiologists in visual analysis of the medical images.

The effect of hair removal and filtering on melanoma detection: a comparative deep learning study with AlexNet CNN.

Melanoma is the most aggressive and prevalent form of skin cancer globally, with a higher incidence in men and individuals with fair skin. Early detection of melanoma is essential for the successful treatment and prevention of metastasis. In this context, deep learning methods, distinguished by their ability to perform automated and detailed analysis, extracting melanoma-specific features, have emerged. These approaches excel in performing large-scale analysis, optimizing time, and providing accurate diagnoses, contributing to timely treatments compared to conventional diagnostic methods. The present study offers a methodology to assess the effectiveness of an AlexNet-based convolutional neural network (CNN) in identifying early-stage melanomas. The model is trained on a balanced dataset of 10,605 dermoscopic images, and on modified datasets where hair, a potential obstructive factor, was detected and removed allowing for an assessment of how hair removal affects the model's overall performance. To perform hair removal, we propose a morphological algorithm combined with different filtering techniques for comparison: Fourier, Wavelet, average blur, and low-pass filters. The model is evaluated through 10-fold cross-validation and the metrics of accuracy, recall, precision, and the F1 score. The results demonstrate that the proposed model performs the best for the dataset where we implemented both a Wavelet filter and hair removal algorithm. It has an accuracy of 91.30%, a recall of 87%, a precision of 95.19%, and an F1 score of 90.91%.

FEATURES OF THE MODIFICATION OF THE INCEPTIONRESNETV2 ARCHITECTURE AND THE CREATION OF A DIAGNOSTIC SYSTEM FOR DETERMINING THE DEGREE OF DAMAGE TO RETINAL VESSELS

Diabetic retinopathy is a retinal disease caused by diabetes. The progression of this disease can lead to blindness. Every year, the number of patients with this disease increases. Diabetic retinal damage can be slowed if it is diagnosed early. The article describes the features of the creation of a neural network model and the development of a system with high accuracy rates for the recognition of diabetic retinopathy. The advantages of the InceptionResNetv2 convolutional neural network architecture are considered. This network uses residual connections that help facilitate the learning process. InceptionResNetv2 uses different methods to reduce the dimensionality of the feature map, making it more economical in terms of memory and computation. This model has a number of advantages compared to other networks. InceptionResNetv2 can be applied to blood vessel segmentation in eye images with different resolutions. In the study, modification of InceptionResNetv2 was carried out. The use of additional MaxPooling and Dense layers improved the speed and accuracy of the InceptionResNetv2 convolutional neural network. The Dropout layer is effectively used to prevent overtraining. The system for determining the degree of retinal damage of diabetic origin is implemented in the Python programming language. Model layers are built using the Keras library. Images from the set of EyePacs were processed by methods of cropping the black frames with a Gaussian blur filter and minimizing the effect of changing the position of the images. During the research, it was found that 21 epochs are needed to achieve maximum accuracy. The program calculates the probability of an image belonging to a certain class with high accuracy. The recognition accuracy rate for class 1 was 98.6%, for class 2 - 98.5%, for class 3 - 98.3%, for class 4 - 98.15%, for class 5 - 98.1%.

Abstract 2584: Decoding colon cancer recurrence: Unveiling accurate predictions with attention-guided deep neural networks on histopathological whole slide images

Abstract Background: Up to 40% of colon cancer patients are at high risk of cancer recurrence, yet accurate and timely prediction tools are lacking. Leveraging whole slide images (WSIs) and deep learning models, we aimed to develop precise algorithm for prediction of colon cancer recurrence. Thus, enables risk stratification for optimized therapeutic interventions and improved health outcomes. Design: We designed an attention-based deep learning model for predicting colon cancer recurrence on paraffin-embedded, hematoxylin and eosin-stained colon tissue biopsies of digital slides. The WSIs were downloaded from the TCGA dataset, and preprocessed for color normalization, tissue segmentation, tilling, and histopathological features extraction. The entire dataset was then labeled based on cancer recurrence status, and divided into training (70%), validation (15%), and testing sets (15%). The model's performance was then evaluated by standard evaluation metrics, including receiver operating characteristic Area Under the Curve (AUC) and accuracy, on the training, validation, and testing datasets, where interpretability attention heatmaps were applied to gain insights into specific histological features and patterns involved in the model's decision-making process. The study is supported by the NIH-NCI-T32 for Next Generation Pathologists Program at our institution. Results: A total of 350 WSIs were included and labeled into two classes: post-therapeutic colon cancer recurrence and no recurrence. The tissue segmentation process involved converting RGB images to HSV color space, applying a median blur filter (kernel size: 7), and performing thresholding using Otsu's method. The extracted features were utilized to train and construct a Clustering-constrained Attention Multiple Instance Learning (CLAM) model. The model demonstrated consistent performance across the validation and testing datasets, achieving an AUC of 0.85 with an accuracy of 0.83 on the validation set, and an identical AUC of 0.85 with an accuracy of 0.83 on the testing set, indicating the model's robust ability to identify patients at risk of cancer recurrence. Conclusion: The study demonstrates the strong performance of our deep learning model in accurately identifying patients at risk for colon cancer recurrence based on H&E WSIs. This capability paves the way for optimizing therapeutic interventions and implementing effective surveillance strategies. Consequently, highlighting the crucial role of pathologists in collaborating with the oncologist for optimizing the management of colon cancer care in the era of personalized medicine. Citation Format: Mohammad K. Alexanderani, Mohamed Omar, Matthew Greenblatt, Ethel Cesarman, Maria T. Diaz-Meco, Jorge Moscat, Luigi Marchionni. Decoding colon cancer recurrence: Unveiling accurate predictions with attention-guided deep neural networks on histopathological whole slide images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2584.

Improvement of crack detectivity for noisy concrete surface by machine learning methods and infrared images

In order to maintain serviceability and reliability of concrete structures, it is essential to assess their condition as concrete structures deteriorate in time. Cracks develop in concrete due to several reasons such as severe loading, environmental effects, chemical effects etc. and cause durability loss in the structure which may also lead to loss of stability. In this research, crack detection is realized by machine learning and an infrared image. The effects of infrared images on crack detection are confirmed by random forest algorithm to select useful explanatory features. Selected features are applied to random forest algorithm and neural network algorithm. Effective filters are selected as a feature selection technique to improve the accuracy. Crack detection is also conducted by U-Net with RGB and infrared images, and the detection characteristics are compared to conventional methods. The performance of two conventional machine learning methods, random forest and neural network, are evaluated based on F1 score and false positives. Applying selected features improves the accuracy of the crack detection from an infrared image. False positives decreased due to monitoring conditions and camera specifications in the infrared image. The most effective image processing filter is the blur filter for each algorithm. Comparing algorithms for crack detection using selected features, different accuracy values are obtained. U-Net enables more accurate crack detection compared to conventional methods. The number of false positives is reduced compare to conventional method. In the detection results by three algorithms, infrared image affects the balance of false negatives and false positives.

Face Recognition Based Attendance System Using Haar Cascade Algorithm with Histogram Equalization and Median Blur Filter

The attendance system is one of the mandatory activities in teaching and learning activities in an educational environment. Unfortunately, many still use a manual attendance system, which is inefficient, and fraud often occurs by manipulating attendance data. In this research, a face recognition-based attendance system was created. The face detection and recognition system use the haar cascade method with image preprocessing, namely histogram equalization and median blur filter. This system can provide output in the form of CSV documents containing attendance data that has been done in real time so that attendance data recording becomes more efficient than before. There is a significant difference in face recognition without and with image preprocessing. Without image preprocessing, the average face recognition accuracy rate is 71.2%. In face recognition with image preprocessing, the face recognition accuracy rate is 91.5%. Therefore, the use of image preprocessing can improve image quality and significantly increase the accuracy of face recognition. In addition, the CSV document is successfully generated automatically after the attendance process is carried out, which is equipped with user data, attendance date, and time when the user makes attendance to avoid fraud in the form of manipulation of attendance data by users.

COEB-SLAM: A Robust VSLAM in Dynamic Environments Combined Object Detection, Epipolar Geometry Constraint, and Blur Filtering

COEB-SLAM: A Robust VSLAM in Dynamic Environments Combined Object Detection, Epipolar Geometry Constraint, and Blur Filtering

Sex estimation from Thai hand radiographs using convolutional neural networks

Manual analysis of hand radiographs for sex estimation is prone to biases and errors. This study addresses the need for automated methods by exploring the potential of Convolutional Neural Networks (CNNs) to accurately identify individual sex from Thai hand radiographs, overcoming limitations in data availability and variable quality. To improve dataset quality, we applied contrast limited adaptive histogram equalization (CLAHE) and Gaussian blur filter techniques to Thai hand radiographs from 385 male and 788 female individuals. We split these images into training, validation, and test sets. We also applied image augmentation techniques to increase the number of radiographs in the training dataset. Seven CNN models were trained, validated, and evaluated on 100 unseen male and female radiographs each. Among these models, the InceptionResNetV2 architecture demonstrated superior performance, achieving an accuracy of 87.50 % and an F1-Score of 86.91 %. Notably, this model utilized information from the 2nd to the 5th metacarpal bone and proximal phalanges in males, and from the 2nd metacarpal bone in females. Our findings provide a solid foundation for sex estimation from Thai hand radiographs, highlighting the power of CNNs in mitigating challenges associated with data quantity and quality. By automating the sex estimation process using CNNs, forensic analysis can benefit from enhanced accuracy and objectivity, enabling faster and more reliable sex assessment. We envisage that future research will build upon these findings to further improve the performance of sex estimation, contributing to advancements in forensic analysis and facilitating more effective utilization of Thai hand radiographs for sex estimation.

Displacement-based structural identification using differentiable physics

Infrastructure is at constant risk of failure due to internal or hidden damage resulting from environmental conditions, variability in building materials, and fabrication errors. These risks can be mitigated by structural health monitoring, but it is difficult to monitor the state of the entire structure using discrete measurements. In this study, we propose a scheme for estimating spatially varying elastic modulus fields of a structural element using differentiable physics to support structural health monitoring efforts. Our approach is evaluated on two numerical model systems, a cantilever beam and a simply supported beam, and systematically evaluated for its sensitivity to initialization, boundary conditions, and a wide variety of spatially varying modulus fields. The results demonstrate the effectiveness of this approach in finding structures with matching displacement responses, while also highlighting its limitations in recovering accurate material properties, as is common in with all inverse methods. To address these limitations, we employ regularization techniques including blur filters and Heaviside projection to mitigate degeneracy and recover more complex fields. This work lays the foundation for more informed decision-making in structural health monitoring, with especially close applications to digital image correlation based approaches.

Mitigating Artifacts in Real-World Video Super-resolution Models

The recurrent structure is a prevalent framework for the task of video super-resolution, which models the temporal dependency between frames via hidden states. When applied to real-world scenarios with unknown and complex degradations, hidden states tend to contain unpleasant artifacts and propagate them to restored frames. In this circumstance, our analyses show that such artifacts can be largely alleviated when the hidden state is replaced with a cleaner counterpart. Based on the observations, we propose a Hidden State Attention (HSA) module to mitigate artifacts in real-world video super-resolution. Specifically, we first adopt various cheap filters to produce a hidden state pool. For example, Gaussian blur filters are for smoothing artifacts while sharpening filters are for enhancing details. To aggregate a new hidden state that contains fewer artifacts from the hidden state pool, we devise a Selective Cross Attention (SCA) module, in which the attention between input features and each hidden state is calculated. Equipped with HSA, our proposed method, namely FastRealVSR, is able to achieve 2x speedup while obtaining better performance than Real-BasicVSR. Codes will be available at https://github.com/TencentARC/FastRealVSR.

Lane Line Detection and Object Scene Segmentation Using Otsu Thresholding and the Fast Hough Transform for Intelligent Vehicles in Complex Road Conditions

An Otsu-threshold- and Canny-edge-detection-based fast Hough transform (FHT) approach to lane detection was proposed to improve the accuracy of lane detection for autonomous vehicle driving. During the last two decades, autonomous vehicles have become very popular, and it is constructive to avoid traffic accidents due to human mistakes. The new generation needs automatic vehicle intelligence. One of the essential functions of a cutting-edge automobile system is lane detection. This study recommended the idea of lane detection through improved (extended) Canny edge detection using a fast Hough transform. The Gaussian blur filter was used to smooth out the image and reduce noise, which could help to improve the edge detection accuracy. An edge detection operator known as the Sobel operator calculated the gradient of the image intensity to identify edges in an image using a convolutional kernel. These techniques were applied in the initial lane detection module to enhance the characteristics of the road lanes, making it easier to detect them in the image. The Hough transform was then used to identify the routes based on the mathematical relationship between the lanes and the vehicle. It did this by converting the image into a polar coordinate system and looking for lines within a specific range of contrasting points. This allowed the algorithm to distinguish between the lanes and other features in the image. After this, the Hough transform was used for lane detection, making it possible to distinguish between left and right lane marking detection extraction; the region of interest (ROI) must be extracted for traditional approaches to work effectively and easily. The proposed methodology was tested on several image sequences. The least-squares fitting in this region was then used to track the lane. The proposed system demonstrated high lane detection in experiments, demonstrating that the identification method performed well regarding reasoning speed and identification accuracy, which considered both accuracy and real-time processing and could satisfy the requirements of lane recognition for lightweight automatic driving systems.

An Optical Approach for Cell Pellet Detection.

Cell-based screening methods are increasingly used in diagnostics and drug development. As a result, various research groups from around the world have been working on this topic to develop methods and algorithms that increase the degree of automation of various measurement techniques. The field of computer vision is becoming increasingly important and has therefore a significant influence on the development of various processes in modern laboratories. In this work we describe an approach for detecting two height information, the phase boundary of a cell pellet and the bottom edge of the tube, and thereby a method for determining the highest point of the topology. The starting point for the development of the method described are cells obtained by various procedures and stabilized by a fixative. Centrifugation of the tube causes the cells to settle to the bottom of the tube, resulting in a cell pellet with a clear phase boundary between the cells and the fixative. For further studies, the supernatant fixative has to be removed without reducing the number of cells. The fixative is to be extracted automatically by a liquid robot, which is only possible by accurately determining the cell pellet height. Due to centrifugation, an uneven topology is formed, which is why the entire phase boundary must be examined to detect the highest point of the cell pellet. For this approach, the tube to be examined, which contains the cells and the fixative, is rotated 360° in defined small steps after centrifugation. During rotation, an image is captured in each step, after which a defined image area is separated from the center of the image and merged into a panoramic image. This produces a panoramic image of the cell topology which represents the complete phase boundary, the boundary located on the outside of the tube. This panoramic image is modified through various image processing steps to extract and detect the phase boundary. Various image processing algorithms from the OpenCV library are used. In the first step, the panoramic image is convolved with a Gaussian blur filter to reduce noise. In the following step, a black and white image is generated by a thresholding process. This black and white image, or binary image, is convolved with a Sobel operator in the x and y directions and the results are superimposed. This overlaid image shows the top edge of the cell pellet and other edges located in the image. A logical exclusion method of the obtained boundaries is used for the detection of the phase boundary. To detect the tube bottom, a multilevel model was trained in advance with an appropriate data set. This model can detect and localize in near real time the tube bottom in an image. By using the two-height information of the different boundaries, phase boundary and tube bottom, the highest point of the cell pellet can be detected. This information is then passed on to a higher-level process so that the liquid robot can approach this point with the pipette tip to remove the excess fixative. By determining the highest point, the probability of being able to remove a larger amount of fixative without reducing the number of cells is highest. This ensures that post-processing studies have the largest possible number of cells available with complete automation.

Deep Neural Network Concept for a Blind Enhancement of Document-Images in the Presence of Multiple Distortions

In this paper, we propose a new convolutional neural network (CNN) architecture for improving document-image quality through decreasing the impact of distortions (i.e., blur, shadows, contrast issues, and noise) contained therein. Indeed, for many document-image processing systems such as OCR (optical character recognition) and document-image classification, the real-world image distortions can significantly degrade the performance of such systems in a way such that they become merely unusable. Therefore, a robust document-image enhancement model is required to preprocess the involved document images. The preprocessor system developed in this paper places “deblurring” and “noise removal and contrast enhancement” in two separate and sequential submodules. In the architecture of those two submodules, three new parts are introduced: (a) the patch-based approach, (b) preprocessing layer involving Gabor and Blur filters, and (c) the approach using residual blocks. Using these last-listed innovations results in a very promising performance when compared to the related works. Indeed, it is demonstrated that even extremely strongly degraded document images that were not previously recognizable by an OCR system can now become well-recognized with a 91.51% character recognition accuracy after the image enhancement preprocessing through our new CNN model.

Crack Detection of Eggshell using Image Processing and Computer Vision

This article presents an eggshell crack inspection using image processing techniques. This approach uses the concept of industrial 4.0 to reduce manual coordination in the egg industry’s manufacturing process. The method started with receiving images from a webcam camera. Then, we rescaled the image to 1147 x 633 for faster computation. Next, divide the image into the red and green channels. The red channel image was converted to grayscale using a Gaussian blur filter with a kernel filter 11 x 11 to reduce noise, followed by turning the image to binary. After that, multiply the binary image with the grayscale of the green channel to remove the background. By that time, a morphological operation was used to enhance the quality of the image. Finally, use the contour matrix to find the area of the object and then build the condition to detect the crack in the eggshell. These techniques of image processing are used to inspect the eggshell crack with a high accuracy of more than 98% as well as the high performance of computing.

Classification of Overlapping Eggs Based on Image Processing

This paper presents a method for classifying the overlapped eggs and counting the number of eggs on the conveyor belt using image processing techniques. The image was acquired by a webcam camera that connected to the computer and then rescaled. The image was then converted to grayscale and noise was reduced using a Gaussian blur filter. Otsu’s Binarization is used to convert the image to binary. The binary image is then subjected to morphological operations. Following that, using the Watershed Algorithm, separate the egg’s overlapped area. Finally, the prepared image is ready to be counted using the contour matrix method. This method independently classifies each egg segmentation and can count up to 18 eggs per frame with a processing time of less than 1 second.

Automated Femoral Landmark Detection from Pelvic X-Rays

Surgical planning requires the identification of anatomical landmarks on medical images pre, intra, and post-procedures. Landmark detection on X-rays is thus an important, though time intensive, step to ensure good surgical outcomes. The operative workflow can be made more efficient by using machine learning approaches to automate the detection of landmarks. These models could also assist with the development of robotic surgical procedures, as robotic approaches to surgery require the robotic device to orient itself to the patient's anatomy. Previously described models are effective, but complex, making it difficult for landmark identification to occur in real-time in a robotic system. We propose a streamlined approach to landmark detection on pelvis radiographs that achieves similar accuracy to gold-standard manual annotations. To train and evaluate the network, 902 pelvic radiographs (382 Outlet view, 520 AP view) were annotated with 22 landmarks and split into training (n = 700), validation (n = 99), and testing (n = 103) set. A U-Net architecture with five encoding layers and five decoding layers was used. Landmark labels were converted into 128 × 128 × 22 heatmap array by placing a Gaussian blur filter (5% of image size) over each landmark in order to generate a single 128 × 128 layer of the array, thus generating the 22 layers of the overall heatmap array with one landmark heatmap on each layer. Each layer was an array of 0 except a circular Gaussian around the area of the landmark. A 256 × 256 array representing the image was fed into the U-Net architecture, which outputted a 128 × 128 × 22 heatmap array. Intersection over union loss was used to train the network. Following training for 100 epochs, the algorithm was able to predict the position of 22 clinically relevant landmarks with an average error of 2.36±.5 mm compared to ground-truth annotations made by trained experts and musculoskeletal radiologists with experience in pelvic scan analysis. The network was able to accurately determine the position of relevant landmarks. It was able to do this despite highly variable contrast and brightness in the dataset. Prediction accuracy was invariant to view (AP: 2.27±.51 mm, Outlet: 2.45±.64 mm). A prediction took 1.13±.07 seconds. The network is able to rapidly provide accurate femoral landmarks on a radiograph of the pelvis. This will allow for the automation of numerous clinically relevant measurements. Additionally, an important avenue of further research is exploring whether automated use of these measurements can aid in the prediction of femur fractures in machine learning models.

Diagnosis of Malaria Infected Blood Cell Digital Images using Deep Convolutional Neural Networks

Automated medical diagnosis is an important topic, especially in detection and classification of diseases. Malaria is one of the most widespread diseases, with more than 200 million cases, according to the 2016 WHO report. Malaria is usually diagnosed using thin and thick blood smears under a microscope. However, proper diagnosis is difficult, especially in poor countries where the disease is most widespread. Therefore, automatic diagnostics helps in identifying the disease through images of red blood cells, with the use of machine learning techniques and digital image processing. This paper presents an accurate model using a Deep Convolutional Neural Network build from scratch. The paper also proposed three CNN models each one trained on the Malaria RBC dataset with different architectures for handling the classification tasks. Furthermore, disadvantage of the traditional method of using transfer learning, and how to control model complexity to achieve better performance was discussed. The dropout regularization technique was used to avoid overfitting problems and minimize validation loss. Applying Data Augmentation technique to avoid the problem of small data in training of proposed models, which is a very common problem in medical dataset. Finally, removing noise in Malaria images using a Median blur filter, and studying how effects of that on training CNN models. According to the classification results, the proposed model achieved better classification results at accuracy 99.22 on the original Malaria RBCs dataset, and it has the best performance comparing with related work.

Knowledge Transferred Fine-Tuning: Convolutional Neural Network Is Born Again With Anti-Aliasing Even in Data-Limited Situations

Anti-aliased convolutional neural networks (CNNs) are models that introduce blur filters to intermediate representations of CNNs to achieve high accuracy in image recognition tasks. A promising way to prepare a new anti-aliased CNN is to introduce blur filters to the intermediate representations of pre-trained (non anti-aliased) CNNs, since many researchers have released them online. Although this scheme can build the new anti-aliased CNN easily, the blur filters drastically degrade the pre-trained representations. Therefore, to take full advantage of the benefits of introducing blur filters, fine-tuning using massive amounts of training data is often required. This can be problematic because the training data is often limited. In such a “data-limited” situation, the fine-tuning does not bring about a high performance because it induces overfitting to the limited training data. To tackle this problem, we propose “knowledge transferred fine-tuning.” Knowledge transfer is a technique that utilizes the representations of a pre-trained model to help ensure generalization in data-limited situations. Inspired by this concept, we transfer knowledge from intermediate representations in a pre-trained CNN to an anti-aliased CNN while fine-tuning. The key idea of our method is to transfer only the essential knowledge for image recognition in the pre-trained CNN using two types of loss: pixel-level loss and global-level loss. The former loss transfers the detailed knowledge from the pre-trained CNN, but this knowledge may contain “aliased” non-essential knowledge. The latter loss, on the other hand, is designed to increase when the pixel-level loss transfers non-essential knowledge while ignoring the essential knowledge, i.e., it penalizes the pixel-level loss. Experimental results demonstrate that the proposed method using just 25 training images per class on ImageNet 2012 can achieve higher accuracy than a conventional pre-trained CNN.

Shielding facial physiological information in video.

With the recent development of non-contact physiological signal detection methods based on videos, it is possible to obtain the physiological parameters through the ordinary video only, such as heart rate and its variability of an individual. Therefore, personal physiological information may be leaked unknowingly with the spread of videos, which may cause privacy or security problems. In this paper a new method is proposed, which can shield physiological information in the video without reducing the video quality significantly. Firstly, the principle of the most widely used physiological signal detection algorithm: remote photoplethysmography (rPPG) was analyzed. Then the region of interest (ROI) of face contain physiological information with high signal to noise ratio was selected. Two physiological information forgery operation: single-channel periodic noise addition with blur filtering and brightness fine-tuning are conducted on the ROIs. Finally, the processed ROI images are merged into video frames to obtain the processed video. Experiments were performed on the VIPL-HR video dataset. The interference efficiencies of the proposed method on two mainly used rPPG methods: Independent Component Analysis (ICA) and Chrominance-based Method (CHROM) are 82.9 % and 84.6 % respectively, which demonstrated the effectiveness of the proposed method.