User-guided Image Editing using Radial Basis Functions

<p>Good image editing tools that modify colors of specified image regions or deform the depicted objects have always been an important part of graphics editors. Manual approaches to this task are too time-consuming, while fully automatic methods are not robust enough. Thus, the ideal editing method should include a combination of manual and automated components. This thesis shows that radial basis functions provide a suitable “engine” for two common image editing problems, where interactivity requires both reasonable performance and fast training. There are many freeform image deformation methods to be used, each having advantages and disadvantages. This thesis explores the use of radial basis functions for freeform image deformation and compares it to a standard approach that uses B-spline warping. Edit propagation is a promising user-guided color editing technique, which, instead of requiring precise selection of the region being edited, accepts color edits as a few brush strokes over an image region and then propagates these edits to the regions with similar appearance. This thesis focuses on an approach to edit propagation, which considers user input as an incomplete set of values of an intended edit function. The approach interpolates between the user input values using radial basis functions to find the edit function for the whole image. While the existing approach applies the user-specified edits to all the regions with similar colors, this thesis presents an extension that propagates the edits more selectively. In addition to color information of each image point, it also takes the surrounding texture into account and better distinguishes different objects, giving the algorithm more information about the user-specified region and making the edit propagation more precise.</p>

User-guided Image Editing using Radial Basis Functions

<p>Good image editing tools that modify colors of specified image regions or deform the depicted objects have always been an important part of graphics editors. Manual approaches to this task are too time-consuming, while fully automatic methods are not robust enough. Thus, the ideal editing method should include a combination of manual and automated components. This thesis shows that radial basis functions provide a suitable “engine” for two common image editing problems, where interactivity requires both reasonable performance and fast training. There are many freeform image deformation methods to be used, each having advantages and disadvantages. This thesis explores the use of radial basis functions for freeform image deformation and compares it to a standard approach that uses B-spline warping. Edit propagation is a promising user-guided color editing technique, which, instead of requiring precise selection of the region being edited, accepts color edits as a few brush strokes over an image region and then propagates these edits to the regions with similar appearance. This thesis focuses on an approach to edit propagation, which considers user input as an incomplete set of values of an intended edit function. The approach interpolates between the user input values using radial basis functions to find the edit function for the whole image. While the existing approach applies the user-specified edits to all the regions with similar colors, this thesis presents an extension that propagates the edits more selectively. In addition to color information of each image point, it also takes the surrounding texture into account and better distinguishes different objects, giving the algorithm more information about the user-specified region and making the edit propagation more precise.</p>

Identifikasi Kemurnian Daging Berbasis Analisis Citra

<p class="Judul2">Daging sapi merupakan salah satu sumber protein hewani yang diperlukan oleh tubuh. Pada tahun 2015 dan 2016 konsumsi daging sapi per kapita sebesar 0,417 kg dan terjadi kenaikan pada tahun 2017 yaitu 12,50 % sebesar 0,469 kg. Sementara harga rata-rata daging sapi di tahun 2015 sebesar Rp 104 747 per kg dan mengalami kenaikan pada tahun 2016 yaitu 8,41 % sebesar Rp 113 555 per kg. Di tahun 2017 kembali terjadi kenaikan yaitu 2,09 % sebesar 115 932 per kg. Berdasarkan sensus penduduk tahun 2010 mendata jumlah penduduk muslim sebesar 207176162 yaitu 87 % dari total penduduk di Indonesia. Kekhawatiran daging halal sangat penting di negara mayoritas muslim. Metode secara konvensional dengan uji laboratorium untuk mendeteksi daging celeng membutuhkan waktu yang relatif lama, tempat khusus, serta biaya yang relatif mahal. Sementara daging yang diwaspadai dicampur dengan daging babi hutan bisa terjadi di berbagai tempat seperti pasar, retailer serta distributor yang sepatutnya bisa dideteksi seketika di tempat tersebut secara cepat. Oleh karena itu, diperlukan sistem yang mudah, cepat, dan mudah dibawa untuk mendeteksi daging sapi murni (tanpa campuran daging lainnya) dalam penelitian ini adalah daging celeng.</p><p class="Paragraf">Paper ini membahas metode deteksi daging campuran berbasis citra menggunakan <em>Convolutional Neural Network </em>(CNN) yang dapat dioperasikan di android. Keunggulan metode ini dapat melakukan proses pembelajaran secara mandiri yaitu ekstraksi citra dan klasifikasi, adapun kemampuan lain yang dimiliki yaitu dapat menangani deformasi gambar seperti translasi, rotasi dan skala. Akurasi yang didapatkan dari metode ini yaitu 94 % untuk mendeteksi daging sapi murni, daging celeng murni, dan daging campuran sapi dan celeng. Sementara presisi untuk celeng, campuran dan sapi yaitu 100 %, 90 % dan 95 %. Selain itu, <em>recall </em>untuk celeng, campuran dan sapi yaitu 85 %, 95 %, dan 97,5 %. Prototipe sistem deteksi yang dikembangkan telah diimplementasikan pada platform android dan diuji pada situasi pencahayaan yang masih terkondisikan. Upaya penyempurnaan ke depan adalah menambah fitur sistem pencahayaan khusus/standar dengan kamera khusus yang memiliki cahaya tambahan yang mengatasi keragaman tingkat pencahayaan di tempat terbuka.</p><p class="Paragraf"> </p><p class="Paragraf"><em><br /></em></p><p class="Paragraf" align="center"><strong><em>Abstract</em></strong></p><p><em>Beef is one of animal protein source that important for human body. In 2015 and 2016 beef consumption per capita was 0.417 kg and it was increasing in 2017 by 12.50 % (i.e., 0.469 kg). While The average price of beef at Rp 104 747 per kg in 2015 and went up by 8,41 % at Rp 113 555 per kg in 2016. In 2017, there was an increase by 2,09 % at Rp 115 932 per kg. The increase of beef price average occurred in 2015 amounting to Rp 104 747 per kg and an increase in 2016 that was 8.41% amounting to Rp 113 555 per kg. Based on the population census in 2010 recorded a Muslim population of 207176162 which is 87% of the total population in Indonesia. The concern of halal (lawful) meat is very critical in the muslim majority country. The conventional method with laboratory testing to detect wild boar meat requires a relatively long time, a special place, and a relatively expensive cost. While meat that is mixed with wild boar can happen in various places such as markets, retailers and distributors which can be detected immediately in that place quickly.Therefore, a system that can be easily, quickly and portably used for detecting pure beef (without other mixed meat) in this study is wild boar. </em></p><p><em>This paper discusses image-based mixed meat detection methods using the Convolutional Neural Network (CNN) that can be operated on android. so the proposed computationally method is Convolutional Neural Network (CNN). The advantages of this method can do the learning process independently, object extraction and classification, while the other capabilities that can handle image deformation such as translation, rotation, and scale. This method yields an overall accuracy of 94% for detecting pure beef, pure wild boar meat, and mixed beef and wild boar. The obtained precision values for wild boar, mixed meat and beef are by 100 %, 90 % and 95 % respectively. Moreover, the values recall for wild boar, mixed meat and beef are by 85 %, 95 % and 97,5 % respectively. The prototype detection system developed has been implemented on the Android platform and tested in a lighting situation that is still conditioned. A future effort to improve is providing special / standard lighting with a special camera that has additional light that can overcome the diversity of levels of exposure in the open areas.</em></p><p> </p><p class="Paragraf"><em><br /></em></p>

Perceptual Properties of the Poisson Effect

When an elastic material (e.g., fabric) is horizontally stretched (or compressed), the material is compressed (or extended) vertically – so-called the Poisson effect. In the different case of the Poisson effect, when an elastic material (e.g., rubber) is vertically squashed, the material is horizontally extended. In both cases, the visual system receives image deformations involving horizontal expansion and vertical compression. How does the brain disentangle the two cases and accurately distinguish stretching from squashing events? Manipulating the relative magnitude of the deformation of a square between horizontal and vertical dimensions in the two-dimensional stimuli, we asked observers to judge the force direction in the stimuli. Specifically, the participants reported whether the square was stretched or squashed. In general, the participant’s judgment was dependent on the relative deformation magnitude. We also checked the anisotropic effect of deformation direction [i.e., horizontal vs. vertical stretching (or squashing)] and found that the participant’s judgment was strongly biased toward horizontal stretching. We also observed that the asymmetric deformation pattern, which indicated the specific context of force direction, was also a strong cue to the force direction judgment. We suggest that the brain judges the force direction in the Poisson effect on the basis of assumptions about the relationship between image deformation and force direction, in addition to the relative image deformation magnitudes between horizontal and vertical dimensions.

Esthetic Evaluation of Facial Soft Tissue Based on Nonrigid Image Deformation

As the time passes by, orthodontics is paying more and more attention to facial aesthetics. However, related research is mostly in the stage of qualitative evaluation without clinical reference of specific soft tissue data. Therefore, by collecting the pre- and post-treatment photographs of 26 adult female patients, this study used image deformation technology to process the photos of a 23-year-old female patient, and two groups of facial photos were established. Then, 22 males and 27 females were selected to conduct an aesthetic evaluation to the original and processed photos by questionnaire survey. Relevant indicators of the corresponding photos were obtained. Group t-test, paired t-test, and nonparametric test were used in data calculation. For patients with high compliance, deep overbite with low angle or average angle are more suitable for fixed appliance with the bite plate, while with high angle are not suitable. The nonprofessionals prefer narrower face, more retracted lip position, and fuller chin than in the actual treatment. Among them, female evaluators prefer narrower lower face than male evaluators do. Male evaluators prefer thicker lips and chins, especially thicker lower lips than female evaluators do. Laypeople prefer narrower face, more retracted lip position, and fuller chin.

A statistical approach to fast nowcasting of lightning potential fields

Thunderstorms and associated hazards like lightning can pose a serious threat to people outside and infrastructure. Thus, very short-term prediction capabilities (called nowcasting) have been developed to capture this threat and aid in decision-making on when to bring people inside for safety reasons. The atmospheric research and operational communities have been developing and using nowcasting methods for decades, but most methods do not rely on formal statistical approaches. A novel and fast statistical approach to nowcasting of lightning threats is presented here that builds upon an integro-difference modeling framework. Inspiration from the heat equation is used to define a redistribution kernel, and a simple linear advection scheme is shown to work well for the lightning prediction example. The model takes only seconds to estimate and nowcast and is competitive with a more complex image deformation approach that is computationally infeasible for very short-term nowcasts.