Defect segmentation for multi-illumination quality control systems

Thanks to recent advancements in image processing and deep learning techniques, visual surface inspection in production lines has become an automated process as long as all the defects are visible in a single or a few images. However, it is often necessary to inspect parts under many different illumination conditions to capture all the defects. Training deep networks to perform this task requires large quantities of annotated data, which are rarely available and cumbersome to obtain. To alleviate this problem, we devised an original augmentation approach that, given a small image collection, generates rotated versions of the images while preserving illumination effects, something that random rotations cannot do. We introduce three real multi-illumination datasets, on which we demonstrate the effectiveness of our illumination preserving rotation approach. Training deep neural architectures with our approach delivers a performance increase of up to 51% in terms of AuPRC score over using standard rotations to perform data augmentation.

Fast and faithful scale-aware image filters

This paper proposes a fast and accurate computational framework for scale-aware image filters. Our framework is based on accurately approximating $$L^{1}$$ L 1 Gaussian convolution with respect to a transformed pixel domain representing geodesic distance on a guidance image manifold in order to recover salient edges in a manner faithful to scale-space theory while removing small image structures. Our framework possesses linear computational complexity with high approximation precision. We examined it numerically in terms of speed, accuracy, and quality compared with conventional methods.

Effects of Color Quantization on JPEG Compression

This paper analyzes the effects of color quantization on standard JPEG compression. Optimized color palettes were used to quantize natural images, using dithering and chroma subsampling as optional. The resulting variations on file size and quantitative quality measures were analyzed. Preliminary results, using a small image database, show that file size suffered an average 20% increase and a concomitant loss in quality was perceived ([Formula: see text]6dB PSNR, [Formula: see text]0.16 SSIM and [Formula: see text]9.6 Butteraugli). Color quantization present itself as an ineffective tool on JPEG compression but if necessarily imposed, on high quality compressed images, it might lead to a negligible increase in data size and quality loss. In addition dithering seems to always decrease JPEG compression ratio.

Method of covert information transfer in still images using a chaotic oscillator.

The subject of the research is the steganographic method of embedding information in digital images. Steganography is able to hide not only the content of information, but also the fact of its existence. The paper presents a method of embedding and extracting information into digital images using a chaotic dynamic system. Chaotic systems are sensitive to certain signals and at the same time immune to noise. These properties allow the use of chaotic systems for embedding information with small image distortions in statistical and visual terms. The methodological basis of the study is the methods of the theory of dynamical systems, mathematical statistics, as well as the theory of image processing. The novelty of the study lies in the development of a new method of embedding information in static images. The author examines in detail the problem of using a chaotic dynamic Duffing system for embedding and extracting information in digital still images. It is shown that the proposed method allows you to embed information in digital images without significant distortion.