Search

In this paper, we present our result of research in video deepfake detection. We built a deepfake detection system to detect whether a video is a deepfake or real. The deepfake detection algorithm still struggle in providing a sufficient accuracy values, especially in challenging deepfake dataset. Our deepfake detection system utilized spatiotemporal feature that extracted using Video Vision Transformer (ViViT). The main contribution of our research is providing a deepfake detection system that based on ViViT architecture and using landmark area images for the input of the system. Our system extracted the feature from a number of spatial features. The spatial feature was extracted using Depthwise Separable Convolution (DSC) block combined with Convolution Block Attention Module (CBAM) from tubelet. The tubelet was a representation of facial landmark area that was extracted from the input video. In our system, we used 25 facial landmark area for an input video. In our experiment we used Celeb-DF version 2 dataset because it is considered to be a challenging deepfake dataset. We conducted augmentation to the dataset, so we obtained 8335 videos for training set, 390 videos for validation set, and 1123 videos for testing set. We trained our deepfake detection system using Adam optimizer, with learning rate of 10–4 and 100 epoch. From the experiment, we obtained the accuracy score of 87.18% and F1 score of 92.52%. We also conducted the ablation study to display the effect of each part of our model to the overall system performance. From this research, we obtained that by using landmark area images, our ViViT based deepfake detection system had a good performance in detecting deepfake videos.

Image captioning is a hot topic that combines a multidiscipline task between computer vision and natural language processing. One of the tasks in the geological field is to make descriptions from the images of geological rocks. The task of a geologist is to write a content description of an image and display it as text that can be used in the future. Interpretation of the object is one of the objectives of the research, which is to traverse the image structures in depth. Shapes, colors, and structures are to be focused on to get the image’s features. The problem faced is how the separable neural network (SNN) and long short-term memory (LSTM) have an impact on the caption that can meet the geologist’s description. SNN is called Visual Attention (VaT), and LSTM is called Semantic Attention (SemAtt) as an architecture of image captioning. The result of the experiment confirms that the accuracy model for captioning gets BLEU- $1=0.908$ , BLEU- $2=0.877$ , BLEU- $3=0.750$ , and BLEU- $4=0.510$ . The evaluation score is compared to those of other evaluators, such as Meteor and RougeL, which get 0.670 and 0.623, respectively. The model confirms that it outperforms the baseline model. Referring to the evaluations, we concluded that the model was able to generate captioned geological rock images that met the geologist’s description. Precision and recall have supported the models in providing the predicted word that is suitable for the image features.

Open-set face recognition on a small dataset with limited image samples per individual poses a significant challenge and is a topic of active research. Therefore, this study investigated the problems of open-set face verification and face identification on a dataset known as the ITB Frontal Profile Limited Dataset (IFPLD), which included only one frontal and one profile image per individual. Various training procedures were used to obtain a more appropriate network embedding for feature representation on the dataset. Transfer learning was employed to improve the performance of the models by fine-tuning the networks using a dataset with properties similar to those of the IFPLD. The results showed that the SimCLR method generated the optimal network embedding for face verification on the Siamese network. The prototypical network with an N-way-k-shot learning scenario where k-1 came from data augmentation outperformed the Siamese network for face identification by a maximum 17.0% accuracy improvement. The transformation from 1-shot learning to k-shot learning is critical for achieving high performance.

Autonomous vehicles rely on the accurate detection and recognition of objects in their surroundings, a critical requirement for safe operation, especially in congested traffic with diverse vehicle types. This study presents a novel dataset collected in various road conditions in Indonesia, and it focuses on the detection and classification of visual objects around autonomous vehicles. Object recognition is achieved through the use of YOLOv7-based deep learning, adapted to identify small, faint, and partially concealed objects. Key enhancements include the integration of a deformable layer and the transition from Non-Maximum Suppression (NMS) to Soft Non-Maximum Suppression (softNMS). The dataset comprises eight predefined custom classes commonly encountered in Indonesian traffic. We collected video data recordings of heavy traffic scenarios featuring a wide range of vehicle types as training and testing data. The object detection model is fine-tuned through transfer learning, with multiple learning configurations explored for comparison. Experimental results demonstrate that deep learning models trained with transfer learning outperform those trained from scratch. Specifically, the modified YOLOv7, referred to as YOLOv7-MOD, incorporates a deformable convolution layer for up-sampling, leading to a remarkable performance of 94.68% in Recall, 96.87% in Precision, and 95.76% in F1-score. The modification resulted in an additional performance increase of 1.05% on average compared to the original model. The findings indicate that YOLOv7-MOD enhances the precision of object detection and recognition compared to the original YOLOv7, making it a promising solution for autonomous vehicle perception systems.

In computer vision, ethnicity classification tasks utilize images containing human faces to extract ethnicity labels. Ethnicity is one of the soft biometric feature categories useful in data analysis for commercial, public, and health sectors. Ethnicity classification begins with face detection as a preprocessing process to determine a human’s presence; then, the feature representation is extracted from the isolated facial image to predict the ethnicity class. This study utilized four handcrafted features (multi-local binary pattern (MLBP), histogram of gradient (HOG), color histogram, and speeded-up-robust-features-based (SURF-based)) as the basis for the generation of a compact-fusion feature. The compact-fusion framework involves optimal feature selection, compact feature extraction, and compact-fusion feature representation. The final feature representation was trained and tested with the SVM One Versus All classifier for ethnicity classification. When it was evaluated in two large datasets, UTKFace and Fair Face, the proposed framework achieved accuracy levels of 89.14%, 82.19%, and 73.87%, respectively, for the UTKFace dataset with four or five classes and the Fair Face dataset with four classes. Furthermore, the compact-fusion feature with a small number of features at 4790, constructed based on conventional handcrafted features, achieved competitive results compared with state-of-the-art methods using a deep-learning-based approach.

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.

The paraphrase generator for citation sentences is used to produce several sentence alternatives to avoid plagiarism. Furthermore, the generation results need to pay attention to semantic similarity and lexical divergence standards. This study proposed the StoPGEN model as an algorithm for generating citation paraphrase sentences with stochastic output. The generation process is guided by an objective function using a simulated annealing algorithm to maintain the properties of semantic similarity and lexical divergence. The objective function is created by combining the two factors that maintain these properties. This study combined METEOR and PINC Scores in a linear weighting function that can be adjusted for its value tendency in one of the matrix functions. The dataset of citation sentences that had been labeled with paraphrases was used to test StoPGEN and other models for comparison. The StoPGEN model, with the citation sentences dataset, produced a BLEU score of 55.37, outperforming the bidirectional LSTM method with a value of 28.93. StoPGEN was also tested using Quora data by changing the language source in the architecture section resulting in a BLEU score of 22.37, outperforming UPSA 18.21. In addition, the qualitative evaluation results of the citation sentence generation based on respondents obtained an acceptance value of 50.80.

Captioning is the process of assembling a description for an image. Previous research on captioning has usually focused on foreground objects. In captioning concepts, there are two main objects for discussion: background object and foreground object. In contrast to the previous image-captioning research, generating captions from the geological images of rocks is more focused on the background of the images. This study proposed image captioning using a convolutional neural network, long short-term memory, and word2vec to generate words from the image. The proposed model was constructed by a convolutional neural network (CNN), long short-term memory (LSTM), and word2vec and gave a dense output of 256 units. To make it properly grammatical, a sequence of predicted words was reconstructed into a sentence by the beam search algorithm with K = 3. An evaluation of the pre-trained baseline model VGG16 and our proposed CNN-A, CNN-B, CNN-C, and CNN-D models used BLEU score methods for the N-gram. The BLEU scores achieved for BLEU-1 using these models were 0.5515, 0.6463, 0.7012, 0.7620, and 0.5620, respectively. BLEU-2 showed scores of 0.6048, 0.6507, 0.7083, 0.8756, and 0.6578, respectively. BLEU-3 performed with scores of 0.6414, 0.6892, 0.7312, 0.8861, and 0.7307, respectively. Finally, BLEU-4 had scores of 0.6526, 0.6504, 0.7345, 0.8250, and 0.7537, respectively. Our CNN-C model outperformed the other models, especially the baseline model. Furthermore, there are several future challenges in studying captions, such as geological sentence structure, geological sentence phrase, and constructing words by a geological tagger.

IFS (Iterated Function Systems) is a method to model fractal object based on affine transformation functions. The star-like object rotation effect in the IFS fractal model could be exhibited by using metamorphical method, as a replacement to the affine rotation method on a non metamorphic animation. The advantage of a metamorphic animation method over the metamorphic animation method is that the object's relative position to the fixed point as an absolute centroid is absolute. Therefore, the rotational effect can be exhibited at any positions. In addition, it can also be combined with rotational effect of the local centroid itself around the absolute centroid as a fixed point by the primitive rotational operation to form an interesting behavior of orbital trajectory Based on the hybrid of both animation methods, the animation simulation could be done on orbital trajectory on a twin stars rotating to each other as a system. Both objects are rotated in the same angular direction, but started in the opposite position around two closely different fixed points. So, the orbital trajectory yielded forms an elliptical path. The two similar objects can be created efficiently by cloning-scaling technique. In general, the animation method can be modeled as an animation framework.

Both non-metamorphic animation and metamorphic animation of objects or multi-objects in IFS fractal form as basic animation method can be implemented by a modified version of the random iteration algorithm as basic algorithm. In this way it is possible to create an attractive animation based on the combination of both animation methods as a generic animation method, as discussed in this paper. The proposed method can be used as an alternative for or as an extension of the existing animation methods in classical geometry, which are restricted to objects with regular forms. The proposed method can be used to produce more attractive media for interactively learning about the behavior of manipulated objects of any form.

Kami menawarkan sebuah optimasi dari fungsi heuristik diferensial terkompresi untuk algoritma Block A*. Optimasi yang ditawarkan berupa optimasi dari segi memori dan performa pencarian yang diukur berdasarkan jumlah simpul yang dikembangkan. Hasil dari optimasi ini menunjukkan bahwa optimasi yang kami tawarkan dapat mengurangi jumlah penggunaan memori walau dengan sedikit penurunan performa pencarian.

Di dalam makalah ini dipresentasikan analisis keamanan dari sebuah usulan algoritma enkripsi citra yang berbasis chaos dan penggunaan teknik enkripsi selektif untuk mengurangi volume komputasi. Algoritma enkripsi menggunakan Arnold Cat Map untuk mengacak citra, selanjutnya teknik enkripsi selektif diterapkan dengan memilih hanya empat bit MSB dari setiap pixel untuk di-XOR-kan dengan keystream yang dibangkitkan dari Logistic Map. Analisis keamanan meliputi analisis ruang kunci, analisis histogram, analisis korelasi, analisis entropi, dan analisis sensitivitas. Secara keseluruhan dapat dinyatakan bahwa analisis keamanan terhadap algoritma enkripsi citra yang ditinjau menunjukkan bahwa algoritma tersebut aman dari berbagai serangan yang bertujuan untuk menemukan kunci atau pixel-pixel di dalam plain-image..

Teka-teki silang (TTS) merupakan salah satu permainan klasik yang masih populer hingga saat ini dan banyak ditemukan di surat kabar dan buku kumpulan TTS. Walaupun bentuk TTS dari masing-masing penerbit berbeda, ada satu persamaan, yakni bentuk papannya yang memiliki simetri, baik lipat maupun putar, sehingga terlihat menarik dan rapi. Untuk membuatnya secara manual tentu membutuhkan kecermatan karena untuk menentukan satu kata sebagai jawaban harus memperhatikan kata-kata yang terhubung dengannya. Di dalam makalah ini dipaparkan aplikasi pembangkit TTS dan sebuah aplikasi permainan TTS berbasis web yang diberi nama Kata Bersilang. Dengan Kata Bersilang, pembuat TTS (designer) hanya perlu merancang papan TTS, sedangkan jawaban dan pertanyaannya akan dibangkitkan secara otomatis oleh Kata Bersilang dengan algoritma backtracking. Permainan TTS pada Kata Bersilang dibuat mirip dengan permainan TTS yang diterbitkan surat kabar di Indonesia. Penjawab TTS (solver) dapat mengisi TTS yang telah dibuat designer selama masih dalam masa pengisian. Setelah masa tersebut berakhir, akan diumumkan pemenangnya.

This paper presents a selective image encryption in frequency domain. At first, the image is transformed into frequency domain with Discrete Cosine Transform (DCT), and then DCT coefficients are scanned in zigzag, and elements of the low-frequency sub-band are extracted. Encryption is performed only on selected elements by scrambling them using 2D chaos map, namely Arnold Cat Map. Next, IDCT is applied to obtain the encrypted image. The encryption algorithm is included in lossy encryption. Experiments on both grayscale images and color images show that the images can be encrypted succesfully. Histograms of the encrypted images differ significantly from histogram of the original images, and the pixels in the encrypted images are not longer correlated.

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).

This paper presents a selective image encryption in frequency domain. At first, the image is transformed into frequency domain with Discrete Cosine Transform (DCT), and then DCT coefficients are scanned in zigzag, and elements of the low-frequency sub-band are extracted. Encryption is performed only on selected elements by scrambling them using 2D chaos map, namely Arnold Cat Map. Next, IDCT is applied to obtain the encrypted image. The encryption algorithm is included in lossy encryption. Experiments on both grayscale images and color images show that the images can be encrypted succesfully. Histograms of the encrypted images differ significantly from histogram of the original images, and the pixels in the encrypted images are not longer correlated.

Teka-teki silang (TTS) merupakan salah satu permainan klasik yang masih populer hingga saat ini dan banyak ditemukan di surat kabar dan buku kumpulan TTS. Walaupun bentuk TTS dari masing-masing penerbit berbeda, ada satu persamaan, yakni bentuk papannya yang memiliki simetri, baik lipat maupun putar, sehingga terlihat menarik dan rapi. Untuk membuatnya secara manual tentu membutuhkan kecermatan karena untuk menentukan satu kata sebagai jawaban harus memperhatikan kata-kata yang terhubung dengannya. Di dalam makalah ini dipaparkan aplikasi pembangkit TTS dan sebuah aplikasi permainan TTS berbasis web yang diberi nama Kata Bersilang. Dengan Kata Bersilang, pembuat TTS (designer) hanya perlu merancang papan TTS, sedangkan jawaban dan pertanyaannya akan dibangkitkan secara otomatis oleh Kata Bersilang dengan algoritma backtracking. Permainan TTS pada Kata Bersilang dibuat mirip dengan permainan TTS yang diterbitkan surat kabar di Indonesia. Penjawab TTS (solver) dapat mengisi TTS yang telah dibuat designer selama masih dalam masa pengisian. Setelah masa tersebut berakhir, akan diumumkan pemenangnya. Kata kunci: pembangkit TTS, permainan TTS, backtracking, Kata Bersilang, designer, solver.

In principal, the computer only recognizes numbers as the representation of a character. Therefore, there are many encoding systems to allocate these numbers although not all characters are covered. In Europe, every single language even needs more than one encoding system. Hence, a new encoding system known as Unicode has been established to overcome this problem. Unicode provides unique id for each different characters which does not depend on platform, program, and language. Unicode standard has been applied in a number of industries, such as Apple, HP, IBM, JustSystem, Microsoft, Oracle, SAP, Sun, Sybase, and Unisys. In addition, language standards and modern information exchanges such as XML, Java, ECMA Script (JavaScript), LDAP, CORBA 3.0, and WML make use of Unicode as an official tool for implementing ISO/IEC 10646. There are four things to do according to Balinese script: the algorithm of transliteration, searching, sorting, and word boundary analysis (spell checking). To verify the truth of algorithm, some applications are made. These applications can run on Linux/Windows OS platform using J2SDK 1.5 and J2ME WTK2 library. The input and output of the algorithm/application are character sequence that is obtained from keyboard punch and external file. This research produces a module or a library which is able to process the Balinese text based on Unicode standard. The output of this research is the ability, skill, and mastering of 1. Unicode standard (21-bit) as a substitution to ASCII (7-bit) and ISO8859-1 (8-bit) as the former default character set in many applications. 2. The Balinese Unicode text processing algorithm. 3. An experience of working with and learning from an international team that consists of the foremost experts in the area: Michael Everson (Ireland), Peter Constable (Microsoft US), I Made Suatjana, and Ida Bagus Adi Sudewa.

Abstract Asymmetric watermarking is the second generation of watermarking scheme which uses different keys for embedding and detecting watermark. Key for embedding is private or secret, but key for detecting can be available publicly and everyone who has the key can detect watermark Watermark detection does not need to be original multimedia data. Detection of watermark is realized using correlation test between public key and multimedia data received. In most of schemes, private key is the watermark itself; public key is public watermark which correlates to the private watermark This paper presents concept of asymmetric watermarking scheme that based on correlation test and reviews some schemes of asymmetric watermarking that have been proposed by researchers. Keywords: asymmetric watermarking, private key, public key, watermark., multimeelia,:correlation.