Showing papers on "Image file formats published in 2019"

Improving embedding efficiency for digital steganography by exploiting similarities between secret and cover images

Abstract: Digital steganography is becoming a common tool for protecting sensitive communications in various applications such as crime/terrorism prevention whereby law enforcing personals need to remotely compare facial images captured at the scene of crime with faces databases of known criminals/suspects; exchanging military maps or surveillance video in hostile environment/situations; privacy preserving in the healthcare systems when storing or exchanging patient’s medical images/records; and prevent bank customers’ accounts/records from being accessed illegally by unauthorized users. Existing digital steganography schemes for embedding secret images in cover image files tend not to exploit various redundancies in the secret image bit-stream to deal with the various conflicting requirements on embedding capacity, stego-image quality, and undetectibility. This paper is concerned with the development of innovative image procedures and data hiding schemes that exploit, as well as increase, similarities between secret image bit-stream and the cover image LSB plane. This will be achieved in two novel steps involving manipulating both the secret and the cover images, prior to embedding, to achieve higher 0:1 ratio in both the secret image bit-stream and the cover image LSB plane. The above two steps strategy has been exploited to use a bit-plane(s) mapping technique, instead of bit-plane(s) replacement to make each cover pixel usable for secret embedding. This paper will demonstrate that this strategy produces stego-images that have minimal distortion, high embedding efficiency, reasonably good stego-image quality and robustness against 3 well-known targeted steganalysis tools.

JPEG XL next-generation image compression architecture and coding tools

Abstract: An update on the JPEG XL standardization effort: JPEG XL is a practical approach focused on scalable web distribution and efficient compression of high-quality images. It will provide various benefits compared to existing image formats: significantly smaller size at equivalent subjective quality; fast, parallelizable decoding and encoding configurations; features such as progressive, lossless, animation, and reversible transcoding of existing JPEG; support for high-quality applications including wide gamut, higher resolution/bit depth/dynamic range, and visually lossless coding. Additionally, a royalty-free baseline is an important goal. The JPEG XL architecture is traditional block-transform coding with upgrades to each component. We describe these components and analyze decoded image quality.

A Novel Visual Medical Image Encryption for Secure Transmission of Authenticated Watermarked Medical Images

Abstract: Secure medical data transmission is the important criteria in tele-medicine. It is helpful for getting the exact medical information at the remote specialist’s side from the local doctor. The medical image watermarking technique is used for secure medical data transmission. Electronic patient record (EPR) is embedded within the original medical image to produce the watermarked medical image for protecting the patient information. The watermarked medical image visually looks like the original medical image. Hence the attacker visually knows the medical image transmission and easily attacks the system. To avoid this situation, image encryption technique is used to change the visual structure of the watermarked medical image. Normal image encryption gives the noise-like or texture-like unknown image format which indicates an existence of some secret information as the encrypted image. To overcome this problem, this paper proposed a visual medical image encryption technique to protect the existence of the watermarked medical image. And to ensure the source authentication, a biometric based authentication is also proposed. The watermarked medical image is encrypted using the proposed visually meaning full image encryption along with the fingerprint image. Experimental results prove that the proposed method yields good results and mitigate the attacker challenges.

Fast Service Migration Method Based on Virtual Machine Technology for MEC

Abstract: In the era of the Internet of Things (IoT), mobile edge computing (MEC) has become an effective solution to meet the energy efficiency and delay requirements of IoT applications. In MEC systems, tasks can be offloaded from lightweight mobile devices to edge nodes that are nearer to the users. To improve user experience, we combine remote loading and redirection to accelerate the service migration. By tracing historic access patterns, the proposed method first generates a loading request list that locates the core codes in the image file of service applications for booting. The core codes are then be prefetched and cached automatically. Furthermore, to avoid the potential UI lagging caused by incomplete service migration, edge nodes can continuously load the remaining codes in the image file. Once the image file is completely migrated, the file will be reconstructed. The running virtual machine (VM) then switches data access to the merged image file. Experiments show that this method can observably reduce the loading time of a VM-based application.

A Spatio-Spectral Hybrid Convolutional Architecture for Hyperspectral Document Authentication

Abstract: Hyperspectral Document Image (HSDI) analysis allows for efficient and accurate differentiation of inks with visually similar color but unique spectral response, which is a crucial step in authentication of documents. Various HSDI based ink discrimination methods are available in the current literature, however, more accurate and robust methods are required to empower document authentication. Contrary to the former ink mismatch detection methods based on spectral features only, we present a novel method based on deep learning that exploits the spectral correlation as well as the spatial context to enhance ink mismatch detection. Spectral responses of the target pixel and its neighboring pixels are organized in an image format and fed to a Convolutional Neural Network (CNN) for classification. The proposed method achieves the highest accuracy among the other ink mismatch detection methods on the UWA Writing Ink Hyperspectral Images database (WIHSI), which demonstrates the effectiveness of deep learning models employing spatio-spectral hybrid features for document authentication. Detailed experimental analysis for selection of appropriate CNN architecture, spatio-spectral data format and training ratio is presented along with a comparison with the previous methods on this subject.

Content-based image retrieval based on combination of texture and colour information extracted in spatial and frequency domains

Abstract: Large amount of data are stored in image format. Image retrieval from bulk databases has become a hot research topic. An alternative method for efficient image retrieval is proposed based on a combination of texture and colour information. The main purpose of this paper is to propose a new content based image retrieval approach using combination of color and texture information in spatial and transform domains jointly.,Various methods are provided for image retrieval, which try to extract the image contents based on texture, colour and shape. The proposed image retrieval method extracts global and local texture and colour information in two spatial and frequency domains. In this way, image is filtered by Gaussian filter, then co-occurrence matrices are made in different directions and the statistical features are extracted. The purpose of this phase is to extract noise-resistant local textures. Then the quantised histogram is produced to extract global colour information in the spatial domain. Also, Gabor filter banks are used to extract local texture features in the frequency domain. After concatenating the extracted features and using the normalised Euclidean criterion, retrieval is performed.,The performance of the proposed method is evaluated based on the precision, recall and run time measures on the Simplicity database. It is compared with many efficient methods of this field. The comparison results showed that the proposed method provides higher precision than many existing methods.,The comparison results showed that the proposed method provides higher precision than many existing methods. Rotation invariant, scale invariant and low sensitivity to noise are some advantages of the proposed method. The run time of the proposed method is within the usual time frame of algorithms in this domain, which indicates that the proposed method can be used online.

Reversible data hiding in JPEG images based on zero coefficients and distortion cost function

Abstract: Recently, reversible data hiding (RDH) in joint photographic experts group (JPEG) images has received a great deal of attention since the JPEG image is one of the most popularly used image formats nowadays. Generally in JPEG image, the quantized discrete cosine transform (DCT) coefficients with the value of zero are far more than those with nonzero coefficients. However, most existing methods are avoided to change these zero coefficients for fear of JPEG file size increase, and the designed distortion cost functions are not perfect. In this paper, we propose a new JPEG RDH method in which the numerous zero coefficients can be modified. Based on an improved distortion cost function, a strategy which can measure the distortion for each coefficient and select coefficient positions which are most suitable for embedding is proposed. With this strategy, zero coefficients are applied to data embedding and the embedding performance is improved. Experimental results have proved that the proposed method can effectively reduce the embedding distortion and increase the embedding capacity while maintaining a relative good file size.

Deep Learning Based Real-time Daily Human Activity Recognition and Its Implementation in a Smartphone

Abstract: A novel method for classifying and identifying human activities in real-time is needed in various human-machine interaction fields. In this paper, a multi-channel motion data collected from a smartphone is structured in a new way and converted to a virtual image. An iOS application software was developed to record and stream motion data and to recognize real-time activities. The time series data of an accelerometer and gyroscope motion sensors are structured into 14x60 virtual image. Similarly, their respective amplitudes of 1 dimensional DFT (Discrete Fourier Transformation) are organized into 14x60 image format. The resultant data was given to the designed CNN (Convolutional Neural Network) for classification. Both data structuring methods were analyzed and compared yet the time series data structuring showed a better result and attained an accuracy of 99.5%. Additionally, the model was tested for real-time activity recognition in a computer and smartphone and achieved an excellent result.

Dataset for file fragment classification of video file formats.

Abstract: File fragment classification of image file formats is a topic of interest in network forensics. There are a few publicly available datasets of files with image formats. Therewith, there is no public dataset for file fragments of image file formats. So, a big research challenge in file fragment classification of image file formats is to compare the performance of the developed methods over the same datasets. In this study, we present a dataset that contains file fragments of ten image file formats: Bitmap, Better Portable Graphics, Free Lossless Image Format, Graphics Interchange Format, Joint Photographic Experts Group, Joint Photographic Experts Group 2000, Joint Photographic Experts Group Extended Range, Portable Network Graphic, Tagged Image File Format, and Web Picture. Corresponding to each format, the dataset contains the file fragments of image files with different compression settings. For each pair of file format and compression setting, 800 file fragments are provided. Totally, the dataset contains 25,600 file fragments.

Deep Multi-scale Discriminative Networks for Double JPEG Compression Forensics

Abstract: As JPEG is the most widely used image format, the importance of tampering detection for JPEG images in blind forensics is self-evident. In this area, extracting effective statistical characteristics from a JPEG image for classification remains a challenge. Effective features are designed manually in traditional methods, suggesting that extensive labor-consuming research and derivation is required. In this article, we propose a novel image tampering detection method based on deep multi-scale discriminative networks (MSD-Nets). The multi-scale module is designed to automatically extract multiple features from the discrete cosine transform (DCT) coefficient histograms of the JPEG image. This module can capture the characteristic information in different scale spaces. In addition, a discriminative module is also utilized to improve the detection effect of the networks in those difficult situations when the first compression quality (QF1) is higher than the second one (QF2). A special network in this module is designed to distinguish the small statistical difference between authentic and tampered regions in these cases. Finally, a probability map can be obtained and the specific tampering area is located using the last classification results. Extensive experiments demonstrate the superiority of our proposed method in both quantitative and qualitative metrics when compared with state-of-the-art approaches.

High Capacity Lossless Data Hiding in JPEG Bitstream Based on General VLC Mapping

Abstract: JPEG is the most popular image format, which is widely used in our daily life. Therefore, reversible data hiding (RDH) for JPEG images is important. Most of the RDH schemes for JPEG images will cause significant distortions and large file size increments in the marked JPEG image. As a special case of RDH, the lossless data hiding (LDH) technique can keep the visual quality of the marked images no degradation. In this paper, a novel high capacity LDH scheme is proposed. In the JPEG bitstream, not all the variable length codes (VLC) are used to encode image data. By constructing the mapping between the used and unused VLCs, the secret data can be embedded by replacing the used VLC with the unused VLC. Different from the previous schemes, our mapping strategy allows the lengths of unused and used VLCs in a mapping set to be unequal. We present some basic insights into the construction of the mapping relationship. Experimental results show that most of the JPEG images using the proposed scheme obtain smaller file size increments than previous RDH schemes. Furthermore, the proposed scheme can obtain high embedding capacity while keeping the marked JPEG image with no distortion.

Steganographic Techniques Classification According to Image Format

Abstract: In this work, we present a classification of steganographic methods applicable to digital images. We also propose a classification of steganographic methods according to the type of image used. We noticed there are no methods that can be applied to all image formats. Each type of image has its characteristics and each steganographic method operates on a precise colorimetric representation. This classification provides an overview of the techniques used for the steganography of digital images

Adaptive Mesh Texture for Multi-View Appearance Modeling

Abstract: In this paper we report on the representation of appearance information in the context of 3D multi-view shape modeling. Most applications in image based 3D modeling resort to texture maps, a 2D mapping of shape color information into image files. Despite their unquestionable merits, in particular the ability to apply standard image tools, including compression, image textures still suffer from limitations that result from the 2D mapping of information that originally belongs to a 3D structure. This is especially true with 2D texture atlases, a generic 2D mapping for 3D mesh models that introduces discontinuities in the texture space and plagues many 3D appearance algorithms. Moreover, the per-triangle texel density of 2D image textures cannot be individually adjusted to the corresponding pixel observation density without a global change in the atlas mapping function. To address these issues, we propose a new appearance representation for image-based 3D shape modeling, which stores appearance information directly on 3D meshes, rather than a texture atlas. We show this representation to allow for input-adaptive sampling and compression support. Our experiments demonstrate that it outperforms traditional image textures, in multi-view reconstruction contexts, with better visual quality and memory footprint, which makes it a suitable tool when dealing with large amounts of data as with dynamic scene 3D models.

Steganography analysis techniques applied to audio and image files

Abstract: The present work carries out a descriptive analysis of the main steganography techniques used in specific digital media such as audio and image files. For this purpose, a literary review of the domains, methods, and techniques as part of this set was carried out and their functioning, qualities, and weaknesses are identified. Hence, it is concluded that there is a wide relationship between audio and image steganography techniques in their implementation form. Nevertheless, it is determined that LSB is one of the weakest techniques, but the safest and the most robust technique within each type of the presented medium.

Estimation affects of formats and resizing process to the accuracy of convolutional neural network

Abstract: image quality, formatting, resizing and compression processes affect on deep neural network performance. These affects are investigated in many papers. However, signal represented images do not look like images which were captured by a camera, because they were plotted in a computer, which let to omit some noises. So formatting and resizing processes are important parameters that affect on network accuracy. In this work, ECG signal representation in different domains were saved in different image formats and CNN trained on these images. Obtained results were compared and showed that JPG image format best fits for training ECG images on CNN.

FQStat: a parallel architecture for very high-speed assessment of sequencing quality metrics.

Abstract: High throughput DNA/RNA sequencing has revolutionized biological and clinical research. Sequencing is widely used, and generates very large amounts of data, mainly due to reduced cost and advanced technologies. Quickly assessing the quality of giga-to-tera base levels of sequencing data has become a routine but important task. Identification and elimination of low-quality sequence data is crucial for reliability of downstream analysis results. There is a need for a high-speed tool that uses optimized parallel programming for batch processing and simply gauges the quality of sequencing data from multiple datasets independent of any other processing steps. FQStat is a stand-alone, platform-independent software tool that assesses the quality of FASTQ files using parallel programming. Based on the machine architecture and input data, FQStat automatically determines the number of cores and the amount of memory to be allocated per file for optimum performance. Our results indicate that in a core-limited case, core assignment overhead exceeds the benefit of additional cores. In a core-unlimited case, there is a saturation point reached in performance by increasingly assigning additional cores per file. We also show that memory allocation per file has a lower priority in performance when compared to the allocation of cores. FQStat’s output is summarized in HTML web page, tab-delimited text file, and high-resolution image formats. FQStat calculates and plots read count, read length, quality score, and high-quality base statistics. FQStat identifies and marks low-quality sequencing data to suggest removal from downstream analysis. We applied FQStat on real sequencing data to optimize performance and to demonstrate its capabilities. We also compared FQStat’s performance to similar quality control (QC) tools that utilize parallel programming and attained improvements in run time. FQStat is a user-friendly tool with a graphical interface that employs a parallel programming architecture and automatically optimizes its performance to generate quality control statistics for sequencing data. Unlike existing tools, these statistics are calculated for multiple datasets and separately at the “lane,” “sample,” and “experiment” level to identify subsets of the samples with low quality, thereby preventing the loss of complete samples when reliable data can still be obtained.

A Residue Number System Based Data Hiding Using Steganography and Cryptography

Abstract: Cryptography and Steganography are two distinct approaches for protected data hiding and diffusion that are widely obtainable. One hides the presence of the message and the other garbles the message. The practices made use of information to cipher or cover their existence correspondingly. Cryptography is the science of using mathematics to encrypt and decrypt data; the data are transformed into some other gibberish form, and then the encrypted data are diffused. While Steganography is the art and science of hiding messages, steganography embeds hidden content in an unexceptional cover media to avoid spy’s suspicion. In steganography the clandestine message embeds in an undisruptive looking cover such as a digital image file and the image file is transmitted. In our proposed method, we use Steganography procedure where secret message is implanted within a hauler image file and the existence of message is hidden from the prowler. To prevent disclosure of contents of the covered file RSA algorithm is used with Steganography to enhance the sturdiness of the system. Usually in practice RSA public and private exponents are chosen to be very large this makes the decryption process slow. And to speed it up we employ the use of Chinese Remainder Theorem which concentrates on modulus calculation. This paper proposes a faster RSA-CRT algorithm for decryption of data. And by employing this technique on RSA algorithm by matching data to an image, there is less chance of an attacker being able to use steganalysis to recover data. Before hiding the data in an image, the application first encrypts it. Keywords: RSA algorithm, cryptography, steganography, LSB method, Chinese Remainder Theorem

Correct use of color for cultural heritage documentation

Abstract: . Accurate color recording is a fundamental feature for proper cultural heritage documentation, cataloging and preservation. However, the methodology used in most cases limits the results since it is based either on perceptual procedures or on the application of digital enhancement techniques only. The objective of this study is to apply a rigorous procedure for the characterization of cameras based on a second-order polynomial model. Trichromatic digital cameras capture color information in the well-known RGB format. Nevertheless, the signal generated by the digital camera is device dependent. By means of the characterization, we establish the relationship between device-dependent RGB values and the tristimulus coordinates defined by the CIE standard colorimetric observer. Once the camera is characterized, users obtain output images in the sRGB space that is independent of the sensor of the camera. We applied the methodology on a set of pictures targeting Levantine rock art motifs in Cova Civil (Castellon, Spain) which is considered part of a UNESCO World Heritage Site. We used raw image files, with different exposure conditions, with raw RGB values captured by the sensor. The outcomes obtained are satisfactory and very promising for proper color documentation in cultural heritage documentation.

Retrieval effectiveness of image search engines

Abstract: The purpose of this study is to explore the retrieval effectiveness of three image search engines (ISE) – Google Images, Yahoo Image Search and Picsearch in terms of their image retrieval capability. It is an effort to carry out a Cranfield experiment to know how efficient the commercial giants in the image search are and how efficient an image specific search engine is.,The keyword search feature of three ISEs – Google images, Yahoo Image Search and Picsearch – was exploited to make search with keyword captions of photos as query terms. Selected top ten images were used to act as a testbed for the study, as images were searched in accordance with features of the test bed. Features to be looked for included size (1200 × 800), format of images (JPEG/JPG) and the rank of the original image retrieved by ISEs under study. To gauge the overall retrieval effectiveness in terms of set standards, only first 50 result hits were checked. Retrieval efficiency of select ISEs were examined with respect to their precision and relative recall.,Yahoo Image Search outscores Google Images and Picsearch both in terms of precision and relative recall. Regarding other criteria – image size, image format and image rank in search results, Google Images is ahead of others.,The study only takes into consideration basic image search feature, i.e. text-based search.,The study implies that image search engines should focus on relevant descriptions. The study evaluated text-based image retrieval facilities and thereby offers a choice to users to select best among the available ISEs for their use.,The study provides an insight into the effectiveness of the three ISEs. The study is one of the few studies to gauge retrieval effectiveness of ISEs. Study also produced key findings that are important for all ISE users and researchers and the Web image search industry. Findings of the study will also prove useful for search engine companies to improve their services.

Image processing device and image processing method

Abstract: To provide an image processing device and an image processing method capable of selecting a special image process whose effect is easy for a user to notice and which is easy to use.SOLUTION: The image processing device comprises: an image processing unit 112 for applying a special image process of imparting a special effect, to a photographed image recorded in an image file; and a display control unit 113 for displaying the photographed image subjected to the special image process on a display element 14.SELECTED DRAWING: Figure 1

ImageProof: Enabling Authentication for Large-Scale Image Retrieval

Abstract: With the explosive growth of online images and the popularity of search engines, a great demand has arisen for small and medium-sized enterprises to build and outsource large-scale image retrieval systems to cloud platforms. While reducing storage and retrieval burdens, enterprises are at risk of facing untrusted cloud service providers. In this paper, we take the first step in studying the problem of query authentication for large-scale image retrieval. Due to the large size of image files, the main challenges are to (i) design efficient authenticated data structures (ADSs) and (ii) balance search, communication, and verification complexities. To address these challenges, we propose two novel ADSs, the Merkle randomized k-d tree and the Merkle inverted index with cuckoo filters, to ensure the integrity of query results in each step of image retrieval. For each ADS, we develop corresponding search and verification algorithms on the basis of a series of systemic design strategies. Furthermore, we put together the ADSs and algorithms to design the final authentication scheme for image retrieval, which we name ImageProof. We also propose several optimization techniques to improve the performance of the proposed ImageProof scheme. Security analysis and extensive experiments are performed to show the robustness and efficiency of ImageProof.

Neural Image Compression and Explanation

Abstract: Explaining the prediction of deep neural networks (DNNs) and semantic image compression are two active research areas of deep learning with a numerous of applications in decision-critical systems, such as surveillance cameras, drones and self-driving cars, where interpretable decision is critical and storage/network bandwidth is limited. In this paper, we propose a novel end-to-end Neural Image Compression and Explanation (NICE) framework that learns to (1) explain the predictions of convolutional neural networks (CNNs), and (2) subsequently compress the input images for efficient storage or transmission. Specifically, NICE generates a sparse mask over an input image by attaching a stochastic binary gate to each pixel of the image, whose parameters are learned through the interaction with the CNN classifier to be explained. The generated mask is able to capture the saliency of each pixel measured by its influence to the final prediction of CNN; it can also be used to produce a mixed-resolution image, where important pixels maintain their original high resolution and insignificant background pixels are subsampled to a low resolution. The produced images achieve a high compression rate (e.g., about 0.6x of original image file size), while retaining a similar classification accuracy. Extensive experiments across multiple image classification benchmarks demonstrate the superior performance of NICE compared to the state-of-the-art methods in terms of explanation quality and semantic image compression rate. Our code is available at: this https URL.

Penerapan Algoritma Secure Hash Algorithm (SHA) Keamanan pada Citra

Abstract: Current information exchange is not only in the form of text, it can be an image or video. All of that can be done using computer network technology in the form of the internet. Through an internet connection, you can connect with many people. Cryptography aims for messages or images that cannot be seen by other parties who have no interest in the information. Messages or secured images can be data stored in secure computer memory or sent through computer networks. And can protect the confidentiality of images from various threats that arise. To maintain data security, SHA-256 is used when transforming data bytes into string hashes. The system built has been able to implement the SHA 256 algorithm which can change the original image file in the form of unknown files and encrypted images that are difficult to know the original image, unless read using an application that has been built. the image when encrypted and used again at the time of decryption with as many as 45 characters and may be in the form of numbers or letters. The SHA 256 process when encrypted in only one round, can be played with decryption.

A Comparative Evaluation of Jpeg Steganography

Abstract: With the dependency on internet and smart gadgets, security of data has become a major concern. Steganography is one of the methods adopted to enhance the security of data by hiding it behind some cover which can be any amongst text, image, audio or video. Images are the most preferred medium to apply steganography as it contains large amount of redundant data. Image steganography hides the data behind the image in a manner that alterations are hardly noticeable to human eye. Images are available in different formats, out of which jpeg images have become the most popular image format due to its small size. These images undergo lossy compression resulting in small sized images, but hiding data behind these images is a challenge. Many algorithms have been suggested till date. This paper intends to give understanding and evolution of different techniques for jpeg steganography. It summarizes various algorithms and techniques used in the past and in present. This paper also analysis's these algorithms based on the basic parameters of image steganography: Undetectability, robustness and embedding capacity.

Firmware version detection method and vulnerability repair rate evaluation method of internet of things device

Abstract: The invention discloses a firmware version detection method and a vulnerability repair rate evaluation method of an Internet of Things device. The detection method comprises the following steps: S1, acquiring a firmware image file; 2, extracting that file system after decompressing the firmware image files, and determining the Web root directory of each file system; 3, searching Web static resources, respectively extracting eigenvalues of each Web static resource in various firmware versions under each device model, and constructing correspond firmware version characteristic tables; S4, obtaining the IP address list and firmware version characteristic table of the required device type, scanning fingerprint, extracting corresponding fingerprint, and identifying the firmware version number of the device; The evaluation method includes obtaining model version information of the target vulnerability, and calculating the repair rate of the target vulnerability according to the result of thedetection method. The invention has the advantages of simple realization method, high detection accuracy and efficiency, and the vulnerability repair rate evaluation without triggering the vulnerability.

Developing a secured image file management system using modified AES

Abstract: Images are a means to share and convey relevant data in today’s digital world. This paper presents an image file management system to provide a platform for distributing and viewing images in a secured manner. The shared image files are stored in the server in an encrypted manner to provide additional security to the owner of the file. A modified AES algorithm using bit permutation was used to encrypt the image files. Based on the experimental result, image files were successfully encrypted in the server and can only be decrypted by the intended recipient of the file providing an efficient and reliable way of exchanging images.

An infinitely scalable dataset of single-polygon grayscale images as a fast test platform for semantic image segmentation

Abstract: Background. Every new semantic image segmentation task requires fine-tuning the segmentation network architecture that is very hard to perform on images of high resolution, which may contain many categories and involve huge computational resources. So, the question is whether it is possible to test segmentation network architectures much faster in order to find optimal solutions that could be imparted to real-world semantic image segmentation tasks.Objective. The goal of the article is to design an infinitely scalable dataset, which could serve as a test platform for semantic image segmentation. The dataset will contain any number of entries of any size required for testing.Methods. A new artificial dataset is designed for semantic image segmentation. The dataset is of grayscale images with the white background. A polygonal object is randomly placed on the background. The polygon edges are black, whereas the polygon body is transparent. Thus, a dataset image is a set of edges of a convex polygon on the white background. The polygon edge is one pixel thick but the transition between the white background and the polygon black edges includes gray pixels in the vicinity of one-pixel edges. Such a noise is an aftermath of the image file format conversion process. The number of edges of the polygon is randomly generated for every next image. The polygon size and position of its center of mass with respect to image margins are randomized as well.Results. A toy dataset of any volume and image size from scratch can be generated. Besides, the dataset generator automatically labels pixels to classes “background” and “polygon”. The dataset does not need augmentation. Eventually, the dataset is infinitely scalable, and it will serve as a fast test platform for segmentation network architectures.Conclusions. The considered examples of using the polygonal dataset confirm its appropriateness and capability of networks trained on it to successfully segment stacks of objects. Additionally, a criterion of early stopping is revealed based on empty image segmentation.

Frame Selection for Producing Recipe with Pictures from an Execution Video of a Recipe

Abstract: In cooking procedure instruction, text format plays an important role in conveying quantitative information accurately, such as time and quantity. On the other hand, image format can smoothly convey qualitative information (e.g., the target food state of a procedure) at a glance. Our goal is to produce multimedia recipes, which have texts and corresponding pictures, for chefs to better understand the procedures. The system takes a procedural text and its unedited execution video as the input and outputs selected frames for instructions in the text. We assume that a frame suits to an instruction when they share key objects. Under this assumption, we extract the information of key objects using named entity recognizer from the text and object detection from the frame, and we convert them into feature vectors and calculate their cosine similarity. To enhance the measurement, we also calculate the scene importance based on the latest changes in object appearance, and aggregate it to the cosine similarity. Finally we align the instruction sequence and the frame sequence using the Viterbi algorithm referring to this suitability and get the frame selection for each instruction. We implemented our method and tested it on a dataset consisting of text recipes and their execution videos. In the experiments we compared the automatic alignment results with those by human annotators. The precision, recall, and F-measure showed that the proposed approach made a steady improvement in this challenging problem of selecting pictures from an unedited video.