Search

Sperm motility measurement using computer assisted sperm analysis (CASA) has been widely accepted as a substitute for manual measurement but still faces several challenges. In the tracking phase, tracking errors caused by detection failure often occur when measuring fresh bull semen. Tracking errors occur for two reasons: (1) the sperm move very fast, which makes them appear blurry, and (2) partial occlusion, which frequently occurs. This study proposes the mean angle of sperm motion and Tracking-Grid to predict the position of the sperm that failed to be detected. The Tracking-Grid has also been found useful in tracking fast-moving sperm. The proposed methods reduce identity switch (ID-switch) and achieve a multi-object tracking overall accuracy (MOTAL) of 73.2. The MOTAL result exhibits 5% less ID-switch and is 15.6 MOTAL points higher than state-of-the-art simple online and real-time tracking with a deep association metric (Deep SORT). The speed achieved is 41.18 frames per second (fps), which is 1.8 times faster than Deep SORT. In sperm motility classification, most researchers use one or several CASA parameters with a static threshold value. Such a method is effective for motile-progressive sperm classification but is less reliable for identifying non-motile-progressive sperm such as vibrating and floating sperm. This study proposes a machine learning-based motility classifier using a support vector machine with three CASA parameters: curvilinear velocity (VCL), straight-line velocity (VSL), and linearity (LIN), which we call the bull sperm progressive motility classifier (BSPMCsvm3casa). Experimental results show that BSPMCsvm3casa’s mean accuracy is 92.08%, which is 2.51–9.67 points higher than other classification methods.

A framework different CNN has been proposed to solve image classification. The power of CNN and the ability to extract demanding features has to be a target for proposed the new ideas. In the geology domain, issues in ascertaining igneous rock from volcanic eruptions often contrast in classification when explored from the location of the rocks. These domain problems must be resolved, contemplating to have consistency and accelerate rock classification. CNN has used to figure out the problem by expanding in multilayer convolution. Besides, parameter tuning has anointed to get the high accuracy to enhance the CNN model. This study has exploited many parameters tuning such as rescaling, cropping, size of inaccurate filter prediction. The exploration has shown that CNN(64,5) achieves a high accuracy of 98.9% and validation carries out accuracy of 81.1%. This study has confirmed that enumerating the tuning parameter on rescaling and cropping does not boost accuracy, even modifying the filter size and stride. Some results have shown still have an inaccuracy class, specifically in the diorite and limestone. The error forecast is 31.7% of 41 predicted diorite images and 30% of 50 predicted limestone images, respectively. (Abstract)

Deepfake technology allows humans to manipulate images and videos using deep learning technology. The results from deepfakes are very difficult to distinguish using ordinary vision. Many algorithms are built to detect deepfake content in images and videos. There are several approaches in deepfake detection, including a visual feature-based approach, a local feature-based approach, a deep feature-based approach and a temporal feature-based approach. The main challenge in developing deepfake detection algorithms is the variety of existing deepfake models in both images and videos. Another challenge is that deepfake technology is still evolving, making deepfake images and videos look more realistic and harder to detect.

Road recognition is one of essential information for determining an Autonomous Vehicle movement. Latest research has shown that machine learning could be used to obtain the information from images. Nevertheless, the system could be improved by effectivity and efficiency. This research proposed finding better feature combinations and using Artificial Neural Network algorithm to build higher accuracy road detection model for better effectivity. Region of Interest module using heuristic method also applied to reduce computation for better efficiency. These three new modules are implemented and combined with road recognition module to become road recognition system. The proposed method performance then tested and compared with the latest research. The experiment results shown that Artificial Neural Network cannot increase the system effectiveness. Nonetheless, with right feature and region of interest module, the proposed system successfully gives better performance. The prototype has accuracy increased from F1-score 0,94 to 0,95 and speed increased from 99 to 112 frames processed per second.

Background and Objective: Object detection is a primary research interest in computer vision. Sperm-cell detection in a densely populated bull semen microscopic observation video presents challenges such as partial occlusion, vast number of objects in a single video frame, tiny size of the object, artifacts, low contrast, and blurry objects because of the rapid movement of the sperm cells. This study proposes an architecture, called DeepSperm, that solves the aforementioned challenges and is more accurate and faster than state-of-the-art architectures. Methods: In the proposed architecture, we use only one detection layer, which is specific for small object detection. For handling overfitting and increasing accuracy, we set a higher network resolution, use a dropout layer, and perform data augmentation on hue, saturation, and exposure. Several hyper-parameters are tuned to achieve better performance. We compare our proposed method with those of a conventional image processing-based object-detection method, you only look once (YOLOv3), and mask region-based convolutional neural network (Mask R-CNN). Results: In our experiment, we achieve 86.91 mAP on the test dataset and a processing speed of 50.3 fps. In comparison with YOLOv3, we achieve an increase of 16.66 mAP point, 3.26 x faster on testing, and 1.4 x faster on training with a small training dataset, which contains 40 video frames. The weights file size was also reduced significantly, with 16.94 x smaller than that of YOLOv3. Moreover, it requires 1.3 x less graphical processing unit (GPU) memory than YOLOv3. Conclusions: This study proposes DeepSperm, which is a simple, effective, and efficient architecture with its hyper-parameters and configuration to detect bull sperm cells robustly in real time. In our experiment, we surpass the state of the art in terms of accuracy, speed, and resource needs., 22 pages, 8 figures, 6 tables, submitted to Computer Methods and Programs in Biomedicine

Steganography is a method of embedding secret messages into a cover file in the form of text, audio, picture or video, so that the message is not suspected by those who are not authorized to open the message. The technique to find out whether the cover media is a stego file or not is steganalysis. In this study, detection of hidden messages focused on MP3 files inserted by the MP3Stego algorithm and Equal Length Entropy Codes Substitution to classify based on algorithms and the estimated length of the message and detect cover files. In conducting this research, it is necessary to know the audio features of MP3, build suitable deep learning methods and the performance of the models that have been produced. The proposed solution for these two problems is to use the QMDCT audio feature and deep learning architecture with Convolutional Neural Network. The results of this study are the best algorithm classification model with an accuracy performance of 91.78% and F1-Score 92.22% and the best classification model for message length estimation has an accuracy performance of 24.16% and F1-Score 21.40%. Thus, the proposal of deep learning architecture is good in classifying algorithms and covers, but still poor in classifying the estimated length of the message.

Convolutional Neural Network (CNN) is one of deep learning methods which is very powerful for solving image processing and computer vision problems. The powerfulness of this method relies on the number and the quality of the training data. The need for numerous and valid annotated training data is crucial. Manual annotation is still the most popular option for obtaining a valid data annotation. There are some manual annotation systems available. Nevertheless, manually annotating a massive amount of data is remarkably tiresome and need plenty of time. This is the problem that we addressed. In this paper, we propose a system to annotate a CNN training data semi-automatically to accelerate the annotation process. We use an extreme case in which the number of objects to annotate inside a video frame is plentiful (334 – 472 objects). We compare the accuracy of its annotation and the speed of the annotation process to the available annotation systems. The result is that our proposed method reduces the average annotation time up to 48.7%.

Extraction of image features done in traditional pixel-based image analysis is not done effectively. This extraction is caused by extraction, which only represents content. A very promising and challenging approach is extracting graphs from images that represent not only content but also the relation of content. The process of graph extraction from an image begins with the segmentation process. The segmentation method using the Minimum Spanning Tree algorithm is one way to divide the image into homogeneous regions. The regions obtained in the segmentation stage are represented as vertices and relations between neighboring regions are represented as edges. This form of representation is called the Region Adjacency Graph (RAG). By assuming that the RAG obtained is a graph feature of the image, then a graph matching process is performed using it. Image graph matching is performed on the artificial images and batik core images using VF2 algorithm and Graph Edit Distance (GED) algorithm. Based on the experiments, the best evaluation results on artificial image data are obtained through the VF2 algorithm with an f-score value of 90.00%. Both precision and recall values for this algorithm have good values, namely 96.24% and 84.52%. Then in the batik core motif, the highest results are achieved through Graph Edit Distance with the GED ≥ 8 parameter. The f-score value obtained is 59.76%, and precision and recall are 47.46% and 60.00%

Video as evidence holds an important place in a court case and therefore the integrity of video must be proven. Various studies had been done in video forensics and most of them is only focused on a certain type of forgery, such as histogram correlation analysis that only focused on detecting temporally forged videos with static background. Improving histogram correlation analysis with foreground detection using Gaussian mixture model makes it possible to detect spatially forged videos and further improves its accuracy in detecting dynamic background video. Applying proposed method will yield 20.83% improved detection's accuracy, 34.42% increased localization's precision, and 39.17% increased localization's re-call. Furthermore, object's definition introduced by foreground detection will also open up new possibility to detect spatially forged video.

Image feature extraction done in image analysis based on the traditional pixel is not extracted effectively. This is due to the extraction that only represents the content. A very promising approach and the challenge are to extract a graph from an image that represents the content and their relationship. The results of graph extraction are obtained from the process of image segmentation. The selection of an appropriate segmentation method from many graph-based image segmentation methods is interesting to be reviewed. The method of super-pixel segmentation is one way to divide the image into regions. The regions obtained in segmentation are represented by vertices and edges represent connections between adjacent regions. This representation is called Region Adjacency Graph (RAG). In this paper, graph extraction of “batik” images has been successfully performed where its representation uses RAG and segmentation uses the Minimum Spanning Tree method. The results of graph extraction have been tested in a simple image retrieval process using the VF2 graph matching method.

Visual cryptography could be used to encrypt secret video so that each participant has their own share in encrypted form. The encrypted video could be played. To recover the original video, all participants have to join their shares. In this paper, we developed a method to encrypt video by using Wang’s scheme. If the secret video has the audio, then the scheme is applied to each frame. The audio could be combined to each share or not. Experiment results show that the scheme could be applied very well to video. The recovered video is same exactly with the original video.

Visual cryptography schemes for color images usually is performed by expanding every pixel in the secret image into some subpixels in the shared images. Wang's schemes are secret image schemes based on XOR operation without expansion of the pixels so that size of shared images are equal to size of the secret images. The schemes are applied to RGB images. In this paper, a method to perform the schemes with minimal computation is proposed. The secret images in RGB color is transformed to YCbCr color model, and computation is performed only on Y component. The comparison of computation between RGB and YCbCr has been done. The results show that by applying the Wang's schemes in YCbCr can minimize the computation without loss of quality.

This paper describes application of a secret colour image sharing scheme based on XOR operation to the animated GIF images. The scheme is performed without expanding every pixel in the secret image so that size of shared images are equal to size of the secret images. Originally the scheme supports RGB images. In this paper we apply the scheme to the animated GIF image. A GIF image consists of a color palette and a matrix which entries (pixel values) refer to the palette row. XOR operation is not performed to RGB palette of GIF image, but to the matrix. The scheme is applied to each frame of the animated GIF image. As a result, each participant has his (or her) own share, each share is an animated GIF image where the frames look like the random images. The experiment shows that the scheme could be applied to the animated GIF image well. The original animated GIF image could be reconstructed exactly.

Digital watermarking is a way to keep the security of digital data such as videos and images. Steganography is a method to hide message or watermark into data in order to keep it secret and avoid attention from others. In this research, an application of video watermarking using steganography or using Discrete Cosine Transform (DCT) method has been developed. DCT is robust method, watermark still be safe even if it's attacked or edited. DCT is often used in compression of JPEG image with transforming coefficients and then doing quantization to reduce the size. The watermark is embedded into quantized coefficients. The coefficients are divided into three frequency: low frequency, middle frequency, and high frequency. Watermark will be embedded using Koch Zhao algorithm where the embedding is performed on middle frequency. After the watermark is successfully embedded, it will be compared to original video's quality using PSNR parameter. The experiment showed that watermarked videos have PSNR $\geq \mathbf{30\ dB}$. The weakness of DCT is against Gaussian Noise and the strength of DCT is against compression. Application built on Android platform because in the present era smartphone technology with camera has been widely used by people and with video watermarking our video will not be owned by others.

Steganography is a science of hiding information. Steganography hides message in certain media such image, text, audio, and video. One of the media is animated GIF. Animated GIF is an image that has a simple animation. Animated GIF can hides message in its frame. Hiding message in animated GIF needs a specific method to embed the message such multibit assignment. Multibit assignment is method to embed message in palette based image and animated GIF is a palette based image. Multibit assignment can hold more than one bit each color in pixel. So that it's possible to embed message in animated GIF using multibit assignment. In this paper propose a method to embed message in animated GIF using multibit assignment method.

Blind steganalysis is a method used to detect whether there is a hidden message in a media without having to know the steganography algorithm behind it. Digital image is converted into features using feature extraction algorithm subtractive pixel adjacency matrix. A model is built based on the resulting features using machine learning method support vector machine. The support vector machine method has different kernel configuration options, which are linear, polynomial, and Gaussian. The model that has been built then undergoes a testing to measure the accuracy performance in message detection and message length estimation. From the model testing, it is obtained that the accuracy in message detection shows good result while the accuracy in message length estimation does not. Highest accuracy is obtained with polynomial kernel.

Automation of visual perception on the machine is the automation of the interpretation of the data generated by the camera sensor like humans. Some automated systems such as the robot and intelligent vehicle relies heavily on an understanding of the environment. The most fundamental understanding of the visual environment is the Machine's ability to identify categories of scenery is still a lot of uncertainty. The uncertainty of the categorization of the scene appears because of the difficulty categories indicated in complex environments, the layout of objects and different scene in the same category. We present a model approach Distribution of the chance appearance of objects and classify them to obtain the semantic meaning of a scene. From the experiments conducted, the model shows high accuracy and using a dataset SUN908 this approach is an effective approach to explore the knowledge-based on the scenes label using large dataset.

This article discusses a new method in detecting the road by using corner adjacent features. Corner is a vertices obtained from every part of vehicle moving from one point to the other, which will be the basic for the road boundary calculation process. The adoption of Lukas Kanade and Melkman algorithm [13] proofs to improve system responsiveness towards the motion of moving object. It is proven that from object reading in the late afternoon with minimum light available, the system able to record corner for 1-minute duration and revealing road boundary masking, in an acceptable level of precision. Nevertheless, system still needs improvement on road condition consisting of trees or any other obstacles producing shadows on the road area, causing the incorrect reading. It is concluded that the proposed system is effectively and efficiently able to read road boundary condition by using a single camera.

This paper discusses a new method in detecting moving objects, which is differ from most of the methods used such as Gaussian Mixture Model, and Haar-Like approach. The focus is on utilizing corner detection and line adjacent detection features through thresholding process creating black and white images to detect the corner of each object in each frame. The process divides a frame length into 4 parts, whereas the first part acted as initiation process of moving object recognition while the rest of the frame functioning as vehicle tracking, speed measurement, and number of vehicles calculation. The initiation process started by identifying corner spots of the moving objects that must be recognized as a single object. The lines surpassing through two points are later identified to determine whether those spots have dark color (0) or light color (1). The moving objects is represented by light color (1) and the walking objects is represented by dark color (0). A group of corner spots, identified and connected by two-point-line equation to be recognized as one unified object by using corner and line adjacent method. The identified vehicle objects can be more easily tracked and identified by the average speed in order to obtain the number of passing vehicles. The research result shows that in the initiation phase, the corner and line adjacent features able to detect moving object and distinguish it with different objects. Furthermore, in tracking phase, system is able to track the vehicle position, measuring the speed and number of vehicles. The system is proven to be able to recognize the moving objects quickly and accurately resulting in the more feasible process of speed measurement and tracking.

This paper presents an asymmetric watermarking method in the DCT domain for still images based on permutation and chaos. This method uses secret watermark as private key and public watermark as public key. The public watermark has a normal distribution with mean = 0 and variance = 1. The secret watermark is obtained by permutating the public watermark according to combination of a part of RC4 algorithm and a logistic map. The watermark is embedded into mid-frequency components of the DCT block for better robustness. The detection process is implemented by correlation test between the public watermark and the received image. Experiments show that the watermarking method was proved to be robust againts some typical image processings (cropping, JPEG compression, resizing, rotation, sharpening, and noising). c

The purpose of this paper is to describe a new approach in road detection. This research uses two detection processes approaches: RGB histogram Filterization and Boundary Classifer, which is different from previous works on road detection. RGB Histogram Filterization processes the reading from the camera in greyscale form and afterward processes them by color segmentation. The last step for this process is determining area between the slopes, which is considered to be the road area. Boundary classification process then employs the RGB indexing on slope ranges, and mapping them into real pictures of roads and its environments. The next process is specifically looking for line boundaries by using Hough-Transform and Canny Edge Detection, and transforms them into binary numbers of ‘0’ and ‘1’. ‘1’ represents road boundaries while ‘0’ represents surrounding area. The coordinate of ‘1’, then mapped by cubic spline to produce connecting line between point ‘1’ coordinates, which in the end produce sharp images on boundaries between road and non-road. This model has proven to be able to detect road conditions and distinguish roads from non-road in a precise way. A test is already conducted for the system by using real-time roads in Bandung, Indonesia. The results are really promising for the road condition on both straight and curved road area.

Energy sustainability is one of the world focuses today. We have built our solution which is called IntelligEnSia (Intelligent Home for Energy Sustainability) that is focused on the prediction analytic using Web and Android technology platforms. In this case, to predict the energy consumption we applied three regression models: simple linear regression, KLM a and KLM b. All models can be applied to predict the next period of energy consumption based on the independent variable of X = day and dependent variables of Y = current, voltage, and power. It can be concluded that KLM a, has the smallest error accuracy among the proposed models. It means that, processing the data of similar period and category in a history, has bigger influence to the prediction value. Based on the testing, it is find out that the biggest error percentage among the models is relied on power, while the smallest is relied on current. These three models are valuable to help the decision maker in creating the better energy management in the city regarding the supply and availability.

In this paper, a fragile watermarking method based on chaos map for image authentication is proposed. The watermark is a binary logo of the same size with the host image. Before embedding, the watermark is encrypted by XOR-ing it with a chaotic image. The chaotic image is generated by using Logistic Map. Next, the least significant bits (LSB) of pixels is used for embedding where the LSB's of the host image are replaced by the encrypted watermark. Authentication is done by extracting the watermark from the watermarked image and then compare it with the original watermark. If found the extracted watermark fragile then it indicates that the image has been altered. Various typical attacks to the watermarked images are carried out to assess the performance of the proposed method. Experiment results show that the proposed method can detect image integrity due to the attacks. For increasing security, the chaos system give sensitivity aspect so that the method is secure from the exhaustive attack.

An interesting of hybrid rotational and revolving simulation of aircraft propellers around a fixed point can be accomplished by the modified version of the shifting centroid method in IFS fractal animation model called the shifting centroid method in double mode. To animate a complex object such as an aircraft with two propellers in fractal model, a simple multi-object of fractal can be used instead of a complicated object which is difficult to be build. The multi-object of fractal can be build by the modified version of random iteration algorithm called partitioned-random iteration algorithm. To rotate many objects as components of a multi-object, the primitive rotation method of affine transformation can be used to a certain object independently.

It is possible to animate each event of two different fractal objects that depend on the event of another fractal object as parts of a single multi-object of fractal. From the multi-object point of view, both the individual events are looked like as a single simultaneous event such as in the rotational simulation of two rotated jet propellers in the double jet engines of aircraft around each local centroid constructed by means of cloning technique and translated relatively to the other fixed point vertically and or horizontally. To accomplish the simulation of shooting aircraft events, also there is a collision detection technique to be proposed. By 2D IFS fractal animation model, at any time the position of fractal objects as components of a multi-object of fractal such as a aircraft and a moving torpedo as an object can be calculated, so the collision between the torpedo object and the aircraft can be simulated by combination of metamorphic animation on multi-object and on object in fractal animation model in 2D point of view. The research result as the contribution for further research in fractal model can be used as an alternative and as an extension of the existing animation method in classical geometry, so it can be used as the attractive media in learning the behavior of the rotated and translated objects in any forms interactively.

Based on a family of transitional 3D IFS code, the attractive metamorphic animation of 3D fractal images between two kinds of 3D fractal images can be conducted. To show an animation which is looked natural, two kinds of 3D fern-like IFS fractal are chosen with small difference of orientation in both x-axis and z-axis directions or horizontally and also in y-axis direction or vertically. There are two ways of animation is conducted, those are with and without spin animations. The spin animation exhibits the three dimensional point of view and the other animation exhibits the accuracy of gradual changing from the start image to the target image without lost of orientation. The 3D IFS code sets as the start and target objects should be in a family of transitional 3D IFS code, otherwise the result of the metamorphic animation will show a strange transition effect.