Edge Adaptation Research Articles

Food image segmentation using edge adaptive based deep-CNNs

PurposeIndian food recognition can be considered as a case of fine-grained type visual recognition, where the several photos of same category generally have significant variability. Therefore, effective segmentation and classification technique is required to identify the particular cuisines and fine-grained analysis. The paper aims to discuss this issue.Design/methodology/approachIn this paper, the authors provided an effective segmentation approach through the proposed edge adaptive (EA)-deep convolutional neural networks (DCNNs) model, where each input images are divided into patches in order to provide much efficient and accurate structural description of data.FindingsEA-DCNNs starts with developing a coarse map of feature that obtained through DCNN, afterwards EA model is applied to construct the final segmented image.Originality/valueThe training model of EA-DCNN consists of pooling, rectified linear unit and convolution, which help convolutional network to optimize the performance of segmentation in a significant extent, which is much practical and relevant in the context of food image segmentation.

Double phase modular steganography with the help of error images

Many steganographic schemes based on popular least significant bit (LSB) embedding have been proposed. The embedding procedure is totally independent of pixel values due to the fact that only the LSB has to be altered. Other methods either randomly increment/decrement the pixel value or use a function to do so. Because of this, the maximum embedding capacity cannot exceed 1 bit per pixel (bpp). State-of-the-art steganographic schemes, such as edge adaptive (EA) and highly undetectable stego (HUGO), aim at such LSB-based approaches with the help of LSB-matching algorithms and, unlike classical steganography, are more concerned about the undetectability level of the stego image rather than the peak signal-to-noise ratio (PSNR) of the stego images. There is a trade-off between detectability and the embedding rate of the stego images. The larger the embedding rate, the higher the chance of detectability, and the opposite is also true. On the other hand, classical steganographic approaches aim for maximum embedding capacity (say 4 bpp) while their main ability is to provide greater PSNR values rather than undetectability. In this work, a novel modular steganographic scheme is proposed in a spatial domain that takes advantage of error images resulting from applying an image quality factor (the same as the ones used in JPEG compression) in order to find the pixels where the secret data could be embedded. We show that our proposed method is less detectable compared to the EA method. The proposed method is almost as undetectable as HUGO. The detectability level is evaluated by the most recent state-of-the-art steganalysis attack, called ensemble classifiers, with second-order subtractive pixel adjacency model features given as their input. In addition, the proposed method can embed up to 4 bpp, yet maintain a high PSNR value.

Double Phase Modular Steganography Using Error Images of Lossy Wavelet Transforms

 Abstract—An error image of Lossy Wavelet Transform differentiates a typical cover image from its wavelet transformed cover image which is resulted from applying a desired image compression factor (ICF) to the respective cover image. The proposed method uses such error images for finding where and to what extent a target pixel in a cover image must hold secret bits. The method is also able to embed maximal embedding rates on gray-scale cover images, up to four bits per pixel (bpp), preserving a human vision-imperceptible PSNR value and less detectability. The detectability level is shown by Ensemble classifiers using second order SPAM as their input features. Additionally, the method embeds minimal rates, up to one bpp, while being roughly as undetectable as state-of-the-art schemes such as HUGO (highly undetectable stego) and EA (edge adaptive).