Recent progress in deep learning, especially deep convolutional neural networks (DCNNs), has led to significant improvement in natural image classification. However, research is still ongoing in the domain of medical image analysis in part due to the shortage of annotated data sets for training DCNNs, the imbalanced number of positive and negative samples, and the difference between medical images and natural images. In this paper, two strategies are proposed to train a DCNN for pathological lymph node image classification. Firstly, the transfer learning strategy is used to deal with the shortage of training samples. Second, a novel exponential loss function is presented for the imbalance in training samples. Four state-of-the-art DCNNs (GoogleNet, ResNet101, Xception, and MobileNetv2) are tested. The experiments demonstrate that the two strategies are effective to improve the performance of pathological lymph node image classification in terms of accuracy and sensitivity with a mean of 0.13% and 1.50%, respectively, for the four DCNNs. In particular, the proposed exponential loss function improved the sensitivity by 3.9% and 4.0% for Xception and ResNet101, respectively.
Currently, there are many papers that have been published on the detection and segmentation of lymph nodes from medical images. However, it is still a challenging problem owing to low contrast with surrounding soft tissues and the variations of lymph node size and shape on computed tomography (CT) images. This is particularly very difficult on low-dose CT of PET/CT acquisitions. In this study, we utilize our previous automatic anatomy recognition (AAR) framework to recognize the thoracic-lymph node stations defined by the International Association for the Study of Lung Cancer (IASLC) lymph node map. The lymph node stations themselves are viewed as anatomic objects and are localized by using a one-shot method in the AAR framework. Two strategies have been taken in this paper for integration into AAR framework. The first is to combine some lymph node stations into composite lymph node stations according to their geometrical nearness. The other is to find the optimal parent (organ or union of organs) as an anchor for each lymph node station based on the recognition error and thereby find an overall optimal hierarchy to arrange anchor organs and lymph node stations. Based on 28 contrast-enhanced thoracic CT image data sets for model building, 12 independent data sets for testing, our results show that thoracic lymph node stations can be localized within 2-3 voxels compared to the ground truth.
Face image symmetry is an important factor affecting the accuracy of automatic face recognition. Selecting high symmetrical face image could improve the performance of the recognition. In this paper, we proposed a novel facial symmetry evaluation scheme based on geometric features, including centroid, singular value, in-plane rotation angle of face and the structural similarity index (SSIM). First, we calculate the value of the four features according to the corresponding formula. Then, we use fuzzy logic algorithm to integrate the value of the four features into a single number which represents the facial symmetry. The proposed method is efficient and can adapt to different recognition methods. Experimental results demonstrate its effectiveness in improving the robustness of face detection and recognition.
Image and video quality measurements are crucial for many applications, such as acquisition, compression, transmission, enhancement, and reproduction. Nowadays, no-reference (NR) image quality assessment (IQA) methods have been drawn extensive attention because it does not need any information of reference images. However, most proposed NR IQA methods are designed only for one or a set of predefined specific distortion types, which are unlikely to generalize for evaluating images distorted with other types of distortions. In order to estimate a wide range of image distortions, in this paper, a novel NR IQA method is proposed which is based on shearlet transform, a new multiscale directional transform with a strong ability to localize distributed discontinuities. The distorted image leads to significant variation in the distributed discontinuities in all directions. Thus, the statistical property of the distorted image is significantly different from that of natural images in shearlet domain. A new model is also proposed to measure this difference. Numerical experiments demonstrate that this new NR IQA method is consistent with subjective assessment, very effective for many well-known types of image distortions and superior to some existing prominent methods.
A digital watermarking technique is a specific branch of steganography, which can be used in various applications,
provides a novel way to solve security problems for multimedia information. In this paper, we proposed a kind of
wavelet domain adaptive image digital watermarking method using chaotic stream encrypt and human eye visual
property. The secret information that can be seen as a watermarking is hidden into a host image, which can be publicly
accessed, so the transportation of the secret information will not attract the attention of illegal receiver. The experimental
results show that the method is invisible and robust against some image processing.
This paper describes a novel approach to multisensor image fusion using a new mathematical transform: the curvelet
transform. The transform has shown promising results over wavelet transform for 2-D signals. Wavelets, though well
suited to point singularities have limitation with orientation selectivity, and therefore, do not represent two-dimensional
singularities (e.g. smooth curves) effectively. Curvelet improves wavelet by incorporating a directional component. This
paper employs the curvelet transform for image fusion. Based on the local energy of direction curvelet subbands, we give
the definition of local band-limited contrast and use it as one of the fusion rules. The local band-limited contrast can
reflect the response of local image features in human visual system truly. When used to image fusion in noiseless
circumstance, it is effective. But in noisy circumstance, it is not always robust. According to the different characteristics
between image features and noise, the local directional energy entropy is proposed. It can distinguish the noise and local
image features. In this paper, the combination of local band-limited contrast and local directional energy entropy is used
as image fusion. Experimental results show that it is robust in noisy and noiseless image fusion system.
There are two ways for transmitting data in a secure manner: Cryptography and steganography. Digital watermarking is a
specific branch of steganography, which can be used in various applications, including covert communication, owner
identification, authentication and copy control. In this paper, we proposed a blind adaptive watermarking algorithm
based on HVS is proper for covert communication. The secret information that can be seen as a watermarking is hidden
into a host image, which can be publicly accessed, so the transportation of the secret information will not attract the
attention of illegal receiver. With our approach, the secret information is embedded in the wavelet domain. By the
background luminance and the texture mask characters of HVS, we divide the wavelet coefficients of carrier image into
different classes. According to the classes of the wavelet coefficients the watermark image is embedded. The result of
our experimental shows that this approach is imperceptible and robust some image processing such as JPEG lossy
compression, cropping, median filtering, grads sharpening, Gaussian white noise attacks and so on.
The existing methods for texture modeling include co-occurrence statistics, filter banks and random fields. However most of these methods lack of capability to characterize the different scale of texture effectively. In this paper, we propose a texture representation which combines local scale feature, amplitude and phase of wavelet modules in multi-scales. The self-similarity of texture is not globally uniform and could be measured in both correlations across the multi-scale and statistical feature within a single-scale. In our approach, the local scale feature is represented by optimal scale obtained through the evolution of wavelet modulus across multi-scales. Then, for all the blocks of the same optimal scale, the statistical measurement of amplitude is extracted to represent the energy within the corresponding frequency
band; the statistical measurement of the phase of modulus is extracted to represent the texture's orientation. Our experiment indicates that, in the proposed texture representation the separability of different texture patterns is larger than the one of the traditional features.
Many advance image processing, like segmentation and recognition, are based on contour extraction which usually lack of ability to allocate edge precisely in the image of heavy noise with low computation burden. For such problem, in this paper, we proposed a new approach of edge detection based on pyramid-structure wavelet transform. In order to suppress noise and keep good continuity of edge, the proposed edge representation considered both inter-correlations across the multi-scales and intra-correlations within the single-scale. The former one is described by point-wise singularity. The later one is described by the magnitude and ratio of wavelet coefficients in different sub-bands. Based on such edge modeling, the edge point allocation is then complemented in wavelet domain by synthesizing the edge information in multi-scales. The experimental results shows that our approaches achieve the pixel-level edge detection with strong resistant against noise due to scattering in water.
Self-similarity features of natural surface play a key role in region segmentation and recognition. Due to long period of
natural evolution, real terrain surface is composed of many
self-similar structures. Consequently, the Self-similarity is
not always so perfect that remains invariable in whole scale space and the traditional single self-similarity parameter can
not represent such abundant self-similarity. In this view, the
self-similarity is not a constant parameter over all scales, but
multi-scale parameters. In order to describe such multi-scale
self-similarities of real surface, firstly we adopt the
Fractional Brownian Motion (FBM) model to estimate the
self-similarity curve of terrain surface. Then the curve is
divided into several linear regions to represent relevant
self-similarities. Based on such regions, we introduce a parameter
called Self-similar Degree (SSD) in the similitude of information entropy. Moreover, the small value of SSD indicates the
more consistent self-similarity. We adopt fifty samples of terrain images and evaluate SSD that represents the multi-scale
self-similarity features for each sample. The samples are clustered by unsupervised fuzzy c mean clustering into various
classes according to SSD and traditional monotone Hurst feature respectively. The measurement for separability of
features shows that the new parameter SSD is an effective feature for terrain classification. Therefore the similarity
feature set that is made up of the monotone Hurst parameter and SSD provides more information than traditional
monotone feature. Consequently, the performance of terrain classification is improved.
In order to improve the quality of image with super-resolution reconstruction, a method based on motion estimation error and edge constraint was proposed. Under the condition of data consistency and amplitude restriction, the motion estimation error was analyzed, with its variance being calculated; meanwhile, in order to suppress the ringing artifacts, edge constraint was adopted and a method based clustering for judging the edge's direction was proposed. The experimental results show that the performance of the this algorithm is better than the traditional linear interpolation and method without considering motion estimation error both in vision effect and peak signal to noise ratio.
Digital watermarking is an efficacious technique to protect the copyright and ownership of digital information. But in the traditional methods of watermarking images, the information of original image will be distorted more or less. Facing this problem, a new watermarking approach, zero-watermarking technique, is proposed. The zero-watermarking approach changes the traditional doings that watermarking is embedded into images, and makes the watermarked image distortion-free. Zero-watermarking can successfully solve the conflict between invisibility and robustness. In this paper, a digital image zero-watermarking method based on discrete wavelet transform and chaotic modulation is proposed.
The zero-watermarking algorithm based on DWT and chaos modulation consists of watermark embedding and detecting processes.
The watermark embedding process is as follow:
First, the original image is decomposed to three-level in wavelet domain. Second, some low frequency wavelet coefficients of original image are selected. The selection of the wavelet coefficients is random by chaotic modulation. Third, the character of coefficients selected is used to construct the character watermark. For each coefficient, in comparison with the adjacent coefficient, we can get the character watermark.
The watermark extracting process is invert process. The location of the coefficients being extracted is also determined by chaotic sequence.
The experimental results show that the watermarking method is invisible and robust against some image processing such as median filtering, JPEG compression, additive Gaussian noise, cropping and rotation attacks and so on. If the initial value of chaos is unknown, the character watermarking can't be extracted correctly.
Weak target inspecting and recovering are very important in IR detecting systems. In this paper, triple correlation peak inspecting techniques (TCPIT) are adopted for the signal processing of IR systems in detecting sub-pixel or point targets. Investigations show that the signal-to-noise ratio (SNR) improvement of approximate 23dB can be obtained with the input peak SNR of 0.84 and the input power SNR of -0.93dB. The triple correlation overlapping sampling technique (TCOST) is advanced for restoring signal waveforms of IR detection systems. Investigations show that signal waveforms can effectively be restored in the low signal-to-noise ratio circumstances using this approach.
Fractal describes the self-similar phenomenon of signal and self-similarity is the most important character of fractal. Pentland provides an excellent explanation of the ruggedness of natural surface. Fractal-based description of image texture has been used effectively in characterization and segmentation of natural scene. A real surface is self-similar over some range of scales, rather than over all scales. That imply self-similarity of a terrain surface is not always so perfect that keep invariable in whole scale space. To describe such self-similarity distribution, a self-similarity curve could be plotted and was divided into several linear regions. We present a new parameter called Self-similarity Degree (SD) in the similitude of information entropy to denote such self-similarity distribution. In addition, one general characterization of self-similarities is result of physical processes. Terrain surface are created by the interactional inogenic and exogenic processes. Hereby, we introduce self-similarity analysis and multifractal singularity spectrum to describe such complex physical field. By the self-similarity analysis and singularity spectrum, the different self-similar structures and the interaction of processes in terrain surface were depicted. Our studies have shown that self-similarity is a relative notion and natural scenes own abundant self-similar structures. Moreover, noises always destroy the self-similarity of original natural surface and change the singularity distribution of original surface.
Recently the protection of digital information has received significant attention and lots of techniques have been proposed. Digital watermarking is an efficacious technique to protect the copyright and ownership of digital information. Since 90' various implementation approaches about digital watermarking have been presented. In this paper, an adaptive blind watermarking algorithm for still images is proposed based on discrete wavelet transform. The wavelet theory mainly used in the multi-resolution analysis and recently has been applied to watermarking technique. In order to be more robust, the embedding strength is decided based on the background luminance and the texture mask characters of HVS, which is adaptive to the carrier image. The experimental results show that the watermarked image has a good quality of image, watermark is imperceptibility, the algorithm is robust against some image processing such as median filtering, JPEG lossy compression, additive Gaussian noise and cropping attacks and so on. Hence our method can be used to protect effectively property right of digital images.
Digital watermarking has been recently proposed as the mean for property right protection of digital products. In this paper we analyze the self-similarities of wavelet transform and present a new approach to embed a digital watermark into an image based on the qualified significant wavelet trees (QSWT) of discrete wavelet transform of the image for the purpose of protecting the copyright of the image. Our studies have shown that the watermarked image has a good quality of image, and such a watermark is difficult to detect and unchangeable without the appropriate user cryptogram.
In this paper, we present a method based on mathematical morphology for digital-hologram synthesis. The principles and operations of mathematical morphology are used for morphologic thansformation of images and for generation of digital image sequence. The sequence of digital images can be used to create kinetic and beatuful hologram which has a good prospect for application in laser anti-counterfeiting field.
Traditional algorithms of 3D surface reconstruction include iteration and linear interpolation and surface fitting technique etc. These methods are mainly suitable for regular shapes and smooth surfaces. For scenes with affluent texture, these methods can't preserve their statistical characteristics. A new method combining wavelet decomposition with fractal interpolation for 3D surface reconstruction is proposed in this paper. For 3D surface reconstruction purposes, the wavelet decomposition extracts strong image self-similar characteristics that can be utilized to three dimension surface reconstruction. With a fractal interpolation, we can generate the practical simulation of three dimension visual surfaces. Our results indicate a good approximation of realistic looking mountainous terrain.
Digital watermarking has been recently proposed as the mean for property right protection of digital products. In this paper we analyze the self-similarities in detail signals of discrete wavelet transform of the image for the purpose of protecting the copyright of the image. The signature embedded using this method is retrievable only by the mean of protected information. Our studies have shown that the watermarked image has a good quality of image, and such a watermark is difficult to detect and unchangeable without the appropriate user crytogram.
With the exceptional development and popularization of laser holography, the application of the holographic anti- counterfeiting identifiers is enlarging gradually. How to improve the quality of holograms for laser holography industries becomes very important. In recent years computer- generated holograms have been investigated intensively because of their wide application range and their advantages in term of flexibility, accuracy, light weight and cost. A method based on Iterated Function System for digital hologram synthesis is proposed in this paper. The method can generate the view angle combining digital holograms by resolving the affine transforms of IFS into many affine transforms of sub- images. The reconstructed sequence of images with multi- channel can be viewed in different angles and look like a kinetic-object. Thus our method provides a new way to laser holography anti-counterfeiting. The generated fractal kinetic holograms have been used in many security holograms.
Conventional method to generate a view angle combining rainbow hologram used a camera to record the sequence of pictures of an object. The object could be a model or a real object. It cost time and money. In recent years, computer-generated holograms have been investigated intensively because of their wide application range and their advantages in term of flexibility, accuracy, light weight and cost. In this paper, a new kind of fractal digital hologram is introduced. The fractal digital holograms are produced with the hyper-complex number model, the fractal hyper-texture model and the self-similar image model based on multi-scale Hurst parameters. The fractal digital holograms have been used in security holograms. In a security hologram, the more complex the model is, the more security the hologram has. The results indicate that fractal digital holograms have a good prospect for application in security holograms.
In this paper, we propose a new kind of fractal kinetic hologram using fractal hypercomplex model for security purposes. The proposed fractal kinetic holograms consist of a sequence of computer-generated fractal images between which there are self-similarity. And on that basis, we can make 'fractal animated holograms.' The results indicate that the model is quite efficient for synthesis of fractal kinetic images. The generated fractal kinetic holograms can be used in the laser anti-counterfeiting field.
In this paper, a kind of self-similar coefficient (SSC) of the range blocks in an image is defined in Fractal Image Compression scheme. It also is proved that the range blocks in fractal images possess short distance piecewise self- similarity (SDPS). A novel edge detect method is proposed based on SSC and SDPS, and the result show that this method can be used to extract edge of fractal image effectively.
A novel image compression method is proposed, in this paper, based on fractal prediction. The original image is decimated into a smaller image which is encoded with fractal method based on the fact that the smaller the size of an image, the shorter the time of encoding. A fractal prediction is obtained by decoding an image, at a same size as the original image, from the fractal codes of the decimated image. The prediction image is subtracted from the original image to arrive at a difference image, which is encoded based on DCT for error correction. Experimental results show that this algorithm is faster than typical fractal image coding methods, and the reconstructed image have good fidelities to the original image at relatively high compression. This method combines the advantages of fractal coding and JPEG coding together.
Proc. SPIE. 3024, Visual Communications and Image Processing '97
KEYWORDS: Image processing algorithms and systems, Signal to noise ratio, Image compression, Image segmentation, Image processing, Computer programming, Image quality, Digital imaging, Fractal analysis, Iterated function systems
Fractal image compression has recently received considerable attention int he literature. In the previously published encoding techniques, an image is usually partitioned into nonoverlapping blocks, and each block is encoded by a self- affirm mapping from a larger block. The fractal code consists of the description of the image partition, along with that of the image transformation defined as the ordered list of block transformation: ((tau) i O <EQ i < N). Each of these block transformations are specified by a set of quantized parameters. With the help of experiment, we discovered the fact that there do exist blocks which have the same transformations are specified by a set of quantized parameters. With the help of experiment, we discovered the fact that there do exist blocks which have the same transformation parameters as the adjacent block transformations. In this paper we propose a region-based transformation that extends the block-based scheme. The concept of the cross search is defined and a search algorithm of finding the region transformations is also given. The results indicate that at the same signal to noise ratio, the region-based system achieves a higher compression ratio over the block-based scheme, and that our algorithm is fast than the block-based system because of less searching.
In recent years, fractal image compression has been paid great attention because of its potential of high compression ratio. In the previously published encoding techniques, an image is usually partitioned into nonoverlapping blocks, and each block is encoded by a self-affirm mapping from a larger block. A high cost of the searching process is generally needed to encode a block. With the help of experiments, we discovered blocks do exist which cannot be well matched with any larger blocks under self-affirm transform. To encode these kinds of blocks with the present fractal encoding method may result in relatively low fidelity on these blocks. In this paper, we propose a hybrid fractal encoding method based on DCT and self- affirm transforms to improve local fidelity and encoding speed. The concept of short distance piecewise self-similarity (SDPS) is defined. Those blocks possessing SDPS are encoded with near-center self-affirm transform method. Other blocks are encoded with quasi-JPEG algorithm. Our method makes use of both the advantages of fractal coding technique, possessing the potential of high compression ratio, and the advantages of JPEG algorithm providing high fidelity at low or medium compression.